Patent Grant
10192138
The present disclosure relates generally to effectively identifying content while limiting the amount of data needed to identify the content. For example, various techniques and systems are provided for identifying content while reducing data density in large datasets.
Managing dense datasets provides significant challenges. For example, there are difficulties in storing, indexing, and managing large amounts of data that is required for certain systems to function. One area in which such problems arise includes systems that search for and identify a closest match between data using reference data stored in large datasets. Storage of the actual data points makes up much of the storage volume in a database.
Certain aspects and features of the present disclosure relate to identifying unknown content. For example, a plurality of vectors can be projected from an origin point. A number of vectors out of the plurality of vectors can be determined that are between a reference data point and an unknown data point. The number of vectors can be used to estimate an angle between a first vector (from the origin point to a reference data point) and a second vector (from the origin point to an unknown data point). A distance between the reference data point and the unknown data point can then be determined. Using the determined distance, candidate data points can be determined from a set of reference data points. The candidate data points can be analyzed to identify the unknown data point.
The techniques described herein allow identification of unknown content, while reducing data density in large datasets. For example, systems and methods are described for improving the efficiency of storing and searching large datasets. The techniques can be applied to any system that harvests and manipulates large volumes of data. Such systems can include, for example, automated content-based searching systems (e.g., automated content recognition for video-related applications or other suitable application), MapReduce systems, Bigtable systems, pattern recognition systems, facial recognition systems, classification systems, computer vision systems, data compression systems, cluster analysis, or any other suitable system.
In some examples, the techniques performed using the systems and methods described herein significantly reduce the amount of data that must be stored in order to search and find relationships between unknown and known data groups. For example, the amount of data that must be stored can be reduced by eliminating the need to store the actual known data points.
According to at least one example, a system is provided for identifying video content being displayed by a display. The system includes one or more processors. The system further includes a non-transitory machine-readable storage medium containing instructions which when executed on the one or more processors, cause the one or more processors to perform operations including: obtaining a plurality of reference video data points; determining a length of a first vector from an origin point to a reference video data point of the plurality of reference video data points; obtaining an unknown video data point associated with video content being presented by a display; determining a length of a second vector from the origin point to the unknown video data point; projecting a plurality of vectors from the origin point; determining a number of the plurality of vectors between the reference video data point and the unknown video data point; estimating an angle between the first vector and the second vector, wherein the angle is estimated using the number of the plurality of vectors; determining a distance between the reference video data point and the unknown video data point, wherein the distance is determined using the estimated angle and the determined lengths of the first vector and the second vector; identifying one or more candidate video data points from the plurality of reference video data points, wherein a candidate video data point is a candidate for matching the unknown video data point, and wherein the one or more candidate video data points are determined based on determined distances between one or more reference video data points and the unknown video data point; and identifying the video content being presented by the display, wherein the video content being presented by the display is identified by comparing the unknown video data point with the one or more candidate video data points.
In another example, a computer-implemented method is provided that includes: obtaining a plurality of reference video data points; determining a length of a first vector from an origin point to a reference video data point of the plurality of reference video data points; obtaining an unknown video data point associated with video content being presented by a display; determining a length of a second vector from the origin point to the unknown video data point; projecting a plurality of vectors from the origin point; determining a number of the plurality of vectors between the reference video data point and the unknown video data point; estimating an angle between the first vector and the second vector, wherein the angle is estimated using the number of the plurality of vectors; determining a distance between the reference video data point and the unknown video data point, wherein the distance is determined using the estimated angle and the determined lengths of the first vector and the second vector; identifying one or more candidate video data points from the plurality of reference video data points, wherein a candidate video data point is a candidate for matching the unknown video data point, and wherein the one or more candidate video data points are determined based on determined distances between one or more reference video data points and the unknown video data point; and identifying the video content being presented by the display, wherein the video content being presented by the display is identified by comparing the unknown video data point with the one or more candidate video data points.
In another example, a computer-program product tangibly embodied in a non-transitory machine-readable storage medium of a computing device may be provided. The computer-program product may include instructions configured to cause one or more data processors to: obtain a plurality of reference video data points; determine a length of a first vector from an origin point to a reference video data point of the plurality of reference video data points; obtain an unknown video data point associated with video content being presented by a display; determine a length of a second vector from the origin point to the unknown video data point; project a plurality of vectors from the origin point; determine a number of the plurality of vectors between the reference video data point and the unknown video data point; estimate an angle between the first vector and the second vector, wherein the angle is estimated using the number of the plurality of vectors; determine a distance between the reference video data point and the unknown video data point, wherein the distance is determined using the estimated angle and the determined lengths of the first vector and the second vector; identify one or more candidate video data points from the plurality of reference video data points, wherein a candidate video data point is a candidate for matching the unknown video data point, and wherein the one or more candidate video data points are determined based on determined distances between one or more reference video data points and the unknown video data point; and identify the video content being presented by the display, wherein the video content being presented by the display is identified by comparing the unknown video data point with the one or more candidate video data points.
In some embodiments, the plurality of reference video data points include video data extracted from one or more video frames.
In some embodiments, the plurality of projected vectors are pseudo-randomly generated.
In some embodiments, determining the number of the plurality of vectors between the reference video data point and the unknown video data point includes: determining whether each vector of the plurality of vectors is to an algebraic right or to an algebraic left of the first vector of the reference video data point; determining whether each vector of the plurality of vectors is to the algebraic right or to the algebraic left of the second vector of the unknown video data point; and determining the number of the plurality of vectors between the reference video data point and the unknown video data point, wherein the number of the plurality of vectors includes vectors to the algebraic left of the first vector and to the algebraic right of the second vector or vectors to the algebraic right of the first vector and to the algebraic left of the second vector.
In some embodiments, the reference video data point is discarded after the length of the first vector of the reference video data point is determined and after each vector of the plurality of vectors is determined to be to the algebraic right or to the algebraic left of the first vector.
In some embodiments, the method, system, and computer-program product described above for identifying video content further includes: storing a first binary value for each vector that is determined to be to the algebraic right of the first vector of the reference video data point; and storing a second binary value for each vector that is determined to be to the algebraic left of the first vector of the reference video data point.
In some embodiments, the method, system, and computer-program product described above for identifying video content further includes: storing a first binary value for each vector that is determined to be to the algebraic right of the second vector of the unknown video data point; and storing a second binary value for each vector that is determined to be to the algebraic left of the second vector of the unknown video data point.
In some embodiments, estimating the angle between the first vector of the reference video data point and the second vector of the unknown video data point includes multiplying a constant by a ratio, wherein the ratio includes the number of the plurality of vectors between the reference video data point and the unknown video data point divided by a total number of the plurality of vectors.
In some embodiments, determining the distance between the reference video data point and the unknown video data point includes performing a Pythagorean identity calculation using the estimated angle and the determined lengths of the first vector and the second vector.
In some embodiments, identifying the video content being presented by the display includes determining a match between the unknown video data point and a candidate video data point, wherein the match is an approximate match based on the candidate video data point being a closest video data point of the one or more candidate video data points to the unknown video data point.
According to at least one other example, a system of identifying one or more unknown data points may be provided that includes one or more processors. The system further includes a non-transitory machine-readable storage medium containing instructions which when executed on the one or more processors, cause the one or more processors to perform operations including: obtaining a plurality of reference data points; determining a length of a first vector from an origin point to a reference data point of the plurality of reference data points; obtaining an unknown data point; determining a length of a second vector from the origin point to the unknown data point; projecting a plurality of vectors from the origin point; determining a number of the plurality of vectors between the reference data point and the unknown data point; estimating an angle between the first vector and the second vector, wherein the angle is estimated using the number of the plurality of vectors; determining a distance between the reference data point and the unknown data point, wherein the distance is determined using the estimated angle and the determined lengths of the first vector and the second vector; and identifying one or more candidate data points from the plurality of reference data points, wherein a candidate data point is a candidate for matching the unknown data point, and wherein the one or more candidate data points are determined based on determined distances between one or more reference data points and the unknown data point.
In another example, a computer-implemented method is provided that includes: obtaining a plurality of reference data points; determining a length of a first vector from an origin point to a reference data point of the plurality of reference data points; obtaining an unknown data point; determining a length of a second vector from the origin point to the unknown data point; projecting a plurality of vectors from the origin point; determining a number of the plurality of vectors between the reference data point and the unknown data point; estimating an angle between the first vector and the second vector, wherein the angle is estimated using the number of the plurality of vectors; determining a distance between the reference data point and the unknown data point, wherein the distance is determined using the estimated angle and the determined lengths of the first vector and the second vector; and identifying one or more candidate data points from the plurality of reference data points, wherein a candidate data point is a candidate for matching the unknown data point, and wherein the one or more candidate data points are determined based on determined distances between one or more reference data points and the unknown data point.
In another example, a computer-program product tangibly embodied in a non-transitory machine-readable storage medium of a television system may be provided. The computer-program product may include instructions configured to cause one or more data processors to: obtain a plurality of reference data points; determine a length of a first vector from an origin point to a reference data point of the plurality of reference data points; obtain an unknown data point; determine a length of a second vector from the origin point to the unknown data point; project a plurality of vectors from the origin point; determine a number of the plurality of vectors between the reference data point and the unknown data point; estimate an angle between the first vector and the second vector, wherein the angle is estimated using the number of the plurality of vectors; determine a distance between the reference data point and the unknown data point, wherein the distance is determined using the estimated angle and the determined lengths of the first vector and the second vector; and identify one or more candidate data points from the plurality of reference data points, wherein a candidate data point is a candidate for matching the unknown data point, and wherein the one or more candidate data points are determined based on determined distances between one or more reference data points and the unknown data point.
In some embodiments, the method, system, and computer-program product described above for identifying one or more unknown data points includes determining a match between the unknown data point and a candidate data point, wherein the match is an approximate match based on the candidate data point being a closest data point of the one or more candidate data points to the unknown data point.
In some embodiments, the plurality of projected vectors are pseudo-randomly generated.
In some embodiments, determining the number of the plurality of vectors between the reference data point and the unknown data point includes: determining whether each vector of the plurality of vectors is to an algebraic right or to an algebraic left of the first vector of the reference data point; determining whether each vector of the plurality of vectors is to the algebraic right or to the algebraic left of the second vector of the unknown data point; and determining the number of the plurality of vectors between the reference data point and the unknown data point, wherein the number of the plurality of vectors includes vectors to the algebraic left of the first vector and to the algebraic right of the second vector or vectors to the algebraic right of the first vector and to the algebraic left of the second vector.
In some embodiments, the reference data point is discarded after the length of the first vector of the reference data point is determined and after each vector of the plurality of vectors is determined to be to the algebraic right or to the algebraic left of the first vector.
In some embodiments, the method, system, and computer-program product described above for identifying one or more unknown data points further includes: storing a first binary value for each vector that is determined to be to the algebraic right of the first vector of the reference data point; and storing a second binary value for each vector that is determined to be to the algebraic left of the first vector of the reference data point.
In some embodiments, the method, system, and computer-program product described above for identifying one or more unknown data points further includes: storing a first binary value for each vector that is determined to be to the algebraic right of the second vector of the unknown data point; and storing a second binary value for each vector that is determined to be to the algebraic left of the second vector of the unknown data point.
In some embodiments, estimating the angle between the first vector of the reference data point and the second vector of the unknown data point includes multiplying a constant by a ratio, wherein the ratio includes the number of the plurality of vectors between the reference data point and the unknown data point divided by a total number of the plurality of vectors.
In some embodiments, determining the distance between the reference data point and the unknown data point includes performing a Pythagorean identity calculation using the estimated angle and the determined lengths of the first vector and the second vector.
This summary is not intended to identify key or essential features of the claimed subject matter, nor is it intended to be used in isolation to determine the scope of the claimed subject matter. The subject matter should be understood by reference to appropriate portions of the entire specification of this patent, any or all drawings, and each claim.
The foregoing, together with other features and embodiments, will become more apparent upon referring to the following specification, claims, and accompanying drawings.
Illustrative embodiments of the present invention are described in detail below with reference to the following drawing figures:
In the following description, for the purposes of explanation, specific details are set forth in order to provide a thorough understanding of embodiments of the invention. However, it will be apparent that various embodiments may be practiced without these specific details. The figures and description are not intended to be restrictive.
The ensuing description provides exemplary embodiments only, and is not intended to limit the scope, applicability, or configuration of the disclosure. Rather, the ensuing description of the exemplary embodiments will provide those skilled in the art with an enabling description for implementing an exemplary embodiment. It should be understood that various changes may be made in the function and arrangement of elements without departing from the spirit and scope of the invention as set forth in the appended claims.
Specific details are given in the following description to provide a thorough understanding of the embodiments. However, it will be understood by one of ordinary skill in the art that the embodiments may be practiced without these specific details. For example, circuits, systems, networks, processes, and other components may be shown as components in block diagram form in order not to obscure the embodiments in unnecessary detail. In other instances, well-known circuits, processes, algorithms, structures, and techniques may be shown without unnecessary detail in order to avoid obscuring the embodiments.
Also, it is noted that individual embodiments may be described as a process which is depicted as a flowchart, a flow diagram, a data flow diagram, a structure diagram, or a block diagram. Although a flowchart may describe the operations as a sequential process, many of the operations can be performed in parallel or concurrently. In addition, the order of the operations may be re-arranged. A process is terminated when its operations are completed, but could have additional steps not included in a figure. A process may correspond to a method, a function, a procedure, a subroutine, a subprogram, etc. When a process corresponds to a function, its termination can correspond to a return of the function to the calling function or the main function.
The term “machine-readable storage medium” or “computer-readable storage medium” includes, but is not limited to, portable or non-portable storage devices, optical storage devices, and various other mediums capable of storing, containing, or carrying instruction(s) and/or data. A machine-readable storage medium or computer-readable storage medium may include a non-transitory medium in which data can be stored and that does not include carrier waves and/or transitory electronic signals propagating wirelessly or over wired connections. Examples of a non-transitory medium may include, but are not limited to, a magnetic disk or tape, optical storage media such as compact disk (CD) or digital versatile disk (DVD), flash memory, memory or memory devices. A computer-program product may include code and/or machine-executable instructions that may represent a procedure, a function, a subprogram, a program, a routine, a subroutine, a module, a software package, a class, or any combination of instructions, data structures, or program statements. A code segment may be coupled to another code segment or a hardware circuit by passing and/or receiving information, data, arguments, parameters, or memory contents. Information, arguments, parameters, data, or other information may be passed, forwarded, or transmitted using any suitable means including memory sharing, message passing, token passing, network transmission, or other transmission technique.
Furthermore, embodiments may be implemented by hardware, software, firmware, middleware, microcode, hardware description languages, or any combination thereof. When implemented in software, firmware, middleware or microcode, the program code or code segments to perform the necessary tasks (e.g., a computer-program product) may be stored in a machine-readable medium. A processor(s) may perform the necessary tasks.
Systems depicted in some of the figures may be provided in various configurations. In some embodiments, the systems may be configured as a distributed system where one or more components of the system are distributed across one or more networks in a cloud computing system.
As described in further detail below, certain aspects and features of the present disclosure relate to identifying unknown data points by comparing the unknown data points to one or more reference data points. The systems and methods described herein improve the efficiency of storing and searching large datasets that are used to identify the unknown data points. For example, the systems and methods allow identification of the unknown data points while reducing the density of the large dataset required to perform the identification. The techniques can be applied to any system that harvests and manipulates large volumes of data. Illustrative examples of these systems include automated content-based searching systems (e.g., automated content recognition for video-related applications or other suitable application), MapReduce systems, Bigtable systems, pattern recognition systems, facial recognition systems, classification systems, computer vision systems, data compression systems, cluster analysis, or any other suitable system. One of ordinary skill in the art will appreciate that the techniques described herein can be applied to any other system that stores data that is compared to unknown data. In the context of automated content recognition (ACR), for example, the systems and methods reduce the amount of data that must be stored in order for a matching system to search and find relationships between unknown and known data groups.
By way of example only and without limitation, some examples described herein use an automated audio and/or video content recognition system for illustrative purposes. However, one of ordinary skill in the art will appreciate that the other systems can use the same techniques.
A significant challenge with ACR systems and other systems that use large volumes of data is managing the amount of data that is required for the system to function. Using a video-based ACR system as one example, one challenge includes attempting to identify a video segment being displayed by a television display in a home among many millions of homes. Another challenge includes the need to build and maintain a database of known video content to serve as a reference to match against. Building and maintaining such a database involves collecting and digesting a vast amount (e.g., hundreds, thousands, or more) of nationally distributed television programs and an even larger amount of local television broadcasts among many other potential content sources. The digesting can be performed using any available technique that reduces the raw data of video or audio into compressed, searchable data (e.g., tokens). With a 24-hour, seven-day-a-week operating schedule and a sliding window of perhaps two weeks of television programming to store, the data volume required to perform ACR builds rapidly. Similar challenges are present with other systems that harvest and manipulate large volumes of data, such as the example systems described above.
The systems and methods described herein allow identification of unknown data points with further reduced datasets than those required using conventional techniques. For example, the amount of data needed to be generated, stored, and compared to search and find relationships between unknown and known data groups is vastly reduced (e.g., by approximately an order of magnitude or other amount depending on the type of system), providing a more efficient technique for storing and indexing the data.
The match request engine 106 can send a communication 124 to a matching engine 112 of the matching server 104. The communication 124 can include a request for the matching engine 112 to identify unknown content. The matching engine 112 can identify the unknown content by matching the content to reference data in a reference database 116. For example, the unknown content can include one or more unknown data points and the reference database 116 can include a plurality of reference data points. In some examples, the unknown content can include unknown video data being presented by a display (for video-based ACR), a search query (for a MapReduce system, a Bigtable system, or other data storage system), an unknown image of a face (for facial recognition), an unknown image of a pattern (for pattern recognition), or any other unknown data that can be matched against a database of reference data. The reference data points can be derived from data received from the data source 118. For example, data points can be extracted from the information provided from the data source 118 and can be indexed and stored in the database 116.
The matching engine 112 can send a request to the candidate determination engine 114 to determine candidate data points from the reference database 116. The candidate data points are reference data points that are a certain determined distance from the unknown data point. The candidate determination engine 114 can return the candidate data points to the matching engine 112. Using the candidate data points, the matching engine 112 can determine a closest reference data point to the unknown data point. For example, as described in more detail below, a path pursuit algorithm can be used to identify the closest reference data point from the candidate data points.
In determining candidate data points 206 for an unknown data point (e.g., unknown data content 202), the candidate determination engine 214 determines a distance between the unknown data point and the reference data points 204 in the reference database. The reference data points that are a certain distance from the unknown data point are identified as the candidate data points 206.
Some examples are described using two-dimensional vector space as an example, but are equally applicable to other vector space dimensions. For example, while the example shown in
Various techniques can be used to determine the distance between data points. For example, one technique of determining a distance between two points A and B, in N-dimensional space, is using the formula:
A.A+B.B−2AB=d^2,
where A is a vector from an origin point (e.g., at point (0,0)) to point A and B is a vector from the origin point to point B. A.A is the dot product of vector A with itself and B.B is the dot product of vector B with itself. The result of A.A and B.B are scalar values.
Another technique of determining a distance between two data points A and B can include using the formula:
A.A/2+B.B/2−A.B=d2/2
Another technique for determining the distance between two or more data points can include using an angle between vectors passing through the data points.
The scalar lengths 406 and 408 can be determined using any suitable technique for determining a length of a vector. One example, shown in
Once the angle 410 and the scalar lengths 406 and 408 are determined, the distance 306 can be determined. For example,
d2=(sin(θ)*B.B)2+(A.A−cosine(θ)*B.B)2,
where θ is the angle 410.
Systems and methods are described herein for determining the distance between data points using vector projections, requiring less data to be stored than the techniques described above. Considering that the dot product A.A is the length of the vector A 402 from the origin O to point A 302, and that the dot product B.B is the length of the vector B 404 from the origin O to point B 304, both of these length values (lengths 406 and 408) can be calculated (e.g., for reference data points) in advance and each length 406 and 408 can be stored as a single number. The only reason to retain an actual point values is for the purposes of calculating the dot product: A.B. An actual unknown data point value should be stored because it is not obtained before run-time when a matching process is performed. For example, an unknown data point is needed during the matching process to compare with the data stored for reference data points. In one example using television content, the matching systems 100 and 200 receive an unknown data point (e.g., data point B 304) when a television sends video data being presented. However, reference data points (e.g., reference data point A 302) can be discarded after they are used to determine information that can then be used to determine the angle between data point vectors (e.g., vectors A 402 and B 404) using projected vectors, as described in more detail below. It is advantageous to discard, and to not to keep, the actual values of reference data points while still being able to calculate the distance between a reference data point (point A) and an unknown data point (point B).
The points A 302 and B 304 have vectors 402 and 404 from an origin (e.g., of (0, 0)) to the respective points. The goal of the candidate determination engine (e.g., candidate determination engine 114 or 214) is to find the distance d 306 between the points A 302 and B 304 in order to identify candidate data points. In some examples, the distance d 306 can be calculated with only the length of vector A 402 (the vector through point A), the length of vector B 404 (the vector through point B), and the angle 410 between vector A 402 and vector B 404.
In some examples, the angle of vector A 402 to the X axis could be stored and then the angle 410 could be calculated, but a disadvantage to this approach would be as the number of dimensions is increased, the system would have to maintain and store angles in every dimension. The result would be a system storing as many numbers defining each point as were previously required (e.g., when all reference data point values are stored).
The systems and methods described herein include generating a number of projections in a defined number of dimensions. For example, a number of vectors can be projected in different directions, such as around the space in which vector A 402 and vector B 404 lie. While the examples described herein use 80 total projected vectors as an illustrative example, one of ordinary skill in the art will appreciate that any number of projections can be generated. The projected vectors can be used to determine the angle between two vectors (e.g., vector A 402 and vector B 404), which can reduce the amount of data needed at run-time during the matching process performed by a matching engine. Using the projection technique, reference data points can be discarded after they are initially used, as described further below.
In one illustrative example briefly describing the technique using projections, 80 regularly distributed vectors may be projected at ten degrees each, in which case theta between the projections is equal to ten. For example, if vector A 402 and vector B 404 are 103 degrees apart, there would be an expected five projections between the vectors A 402 and B 404. It might intuitively seem that there would be 10 projections between the vectors A 402 and B 404, but there are five. For example, projections extend in both directions and a projection projecting into the third quadrant will still be “in between” A and B as far as the angle is concerned. However, for the purpose of the examples discussed herein, each projection can be considered as being only in one quadrant, as this would be close enough for conceptual purposes. In this example, because five of the vectors out of the 80 projected vectors fall between vector A 402 and vector B 404, it can be determined that the angle between vector A 402 and vector B 404 is 10 degrees, as described in more detail below. Once the angle 410 between the two vectors A 402 and B 404 and the lengths of vector A 402 and vector B 404 are determined, trigonometry can then be used to calculate the distance d 306 between the points A 302 and B 304. Further details are provided with respect to
For each reference data point in the reference database, a matching system determines whether each projected vector is to the algebraic right or to the algebraic left of each vector of each reference data point (e.g., vector A 402 of point A 302), such as by calculating the dot product of a vector (from an origin to a reference data point) with a projected vector, as described below with respect to
When an unknown data point is received (e.g., when a video data point is received from a television), the matching system can determine candidate data points from the information stored for the reference data points (e.g., the left and right binary data and vector lengths for the reference data points), for example, by searching for nearest neighbor points in the reference database. The matching system can then determine whether the projected vectors are to the algebraic right or left of a vector for the unknown data point (e.g., by taking the dot product) to get the left and right binary values. The matching system can also calculate the length of the vector of the unknown data point. Projected vectors that fall between a reference data point and an unknown data point can be used to determine an angle between the vector of the reference data point and the vector of the unknown data point.
For example, as shown in
When it comes time to calculate the distance between the unknown data point 304 and the reference data points (including data point A 302), the exclusive OR is calculated between the binary data (for the projections) of each reference data point to the binary data value of the unknown data point. As noted above, the result of the exclusive or between the binary data of the unknown data point (e.g., data point B 304) and a reference data point (e.g., data point A 302) is the number of projections between the unknown data point and the reference data point. Again, the derived angle 910 equals the number of projections between the data points 302 and 304 divided by the number of total projections (80 in this example), multiplied by 180. The derived distance can then be computed by performing a Pythagorean identity calculation using the formula:
d2=(sin(θ)*B.B)2+(A.A−cos(θ)*B.B)2
The dot product B.B represents the length of the vector B 404, and the dot product A.A represents the length of the vector A 402.
Once the distance from the unknown data point B 304 to the reference data points stored in a reference database is determined, candidate data points can be determined, as described previously. The unknown data point B 304 can then be identified by one of the candidate data points. For example, one of the reference data points can be determined as a closest match with the unknown data point B 304. The unknown data point B 304 can then be identified as being the same data as the candidate data point found as the closest match. In some examples, a nearest neighbor approach can be used to identify candidate data points, and a path pursuit algorithm can be used to identify the unknown data point. Details of the nearest neighbor and path pursuit techniques are described with respect to
Using the above-described vector projection technique, less data is needed to determine a distance between a reference data point and an unknown data point. Such a technique can be used to find candidate data points among a plurality of reference data points. As described previously, the length of the vector A 402 (denoted by the dot product A.A) is already known before the unknown data point is received, and thus is not needed at run time to determine the distances for generating candidate data points. The only data needed are the lengths (or distances) of the reference data point vectors and the bits representing the algebraic left and right binary data of projected vectors relative to the reference and unknown data points (which are used to determine the angle between reference and unknown data point vectors). For example, since dot products are calculated for all of the projected vectors with respect to point A 302 (before the matching process) and for all of the projected vectors with respect to point B 304, the matching system can store a bit for each projection for each reference data point and a bit for each projection for the unknown data point, in addition to the vector lengths of the reference data points. Accordingly, when comparing an unknown data point to reference data points to identify candidate data points at run time, the matching system can discard the actual reference data points.
An advantage of the vector projection technique described above is that the exact amount of data saved depends on how many projections are needed to obtain acceptable results. For example, each projection adds one bit to memory use when compared to one data point, so if the number of random projections is a reasonable number, such as 80, an original 75-byte dataset (point) can be replaced with a 10-byte left/right binary structure plus, for example, 2 bytes for the length of the vector for that point, totaling 12 bytes. Such an example provides a savings of 63 bytes for each point in memory. Hence, the vector projection technique provides a highly advantageous reduction in data size when computing large data searches and comparisons, and can be used for many large-scale applications.
The vectors can be projected randomly or pseudo-randomly. Pseudo-random projection includes predetermined projections in a distributed manner. For example, a machine can generate the same projections numerous times, but the projections can appear to be random. Random projections are random, though one skilled in the art would understand that they are actually selected from a normal distribution randomly. Therefore, not all space definitions are created equal. As is known to the skilled person, a Monte Carlo method can be employed to pick a random or pseudo-random projection that is good. Once a suitable, adequate random projection is picked, the random projection can be used for many distance determinations (e.g., to determine angles between many data point vectors), and there is no need to change it unless it is desirable to increase or decrease the number of projections. As this is a probabilistic calculation, the result may not be the correct answer, but will be very close to correct. A skilled person will understand that the result does not need to be “correct,” but only close enough to provide a useful utility to big data applications. In order to measure an adequate random distribution, a direct suitability test is performed, as discussed in more detail below.
An example of a satisfactorily distributed random set of projections is shown in
The suitability of the projections can be determined once by experimentation and kept and used for random or pseudo-random generation. For example, the suitability of the projected vectors can be determined by performing a sufficiency test that tests the distance determination technique described above (using the projections to determine an angle between vectors of a reference point and an unknown point) in comparison to prior system and comparing the results. A goal of 5% error can be used as a threshold to determine the suitability of the randomly projected vectors. One of ordinary skill in the art will appreciate that any threshold percentage can be used, depending on the particular application and required quality desired.
In such a case, another group of projections is generated to attempt to obtain a satisfactory projection. The projected vectors of
Process 1700 is illustrated as a logical flow diagram, the operation of which represent a sequence of operations that can be implemented in hardware, computer instructions, or a combination thereof. In the context of computer instructions, the operations represent computer-executable instructions stored on one or more computer-readable storage media that, when executed by one or more processors, perform the recited operations. Generally, computer-executable instructions include routines, programs, objects, components, data structures, and the like that perform particular functions or implement particular data types. The order in which the operations are described is not intended to be construed as a limitation, and any number of the described operations can be combined in any order and/or in parallel to implement the processes.
Additionally, the process 1700 may be performed under the control of one or more computer systems configured with executable instructions and may be implemented as code (e.g., executable instructions, one or more computer programs, or one or more applications) executing collectively on one or more processors, by hardware, or combinations thereof. As noted above, the code may be stored on a machine-readable storage medium, for example, in the form of a computer program comprising a plurality of instructions executable by one or more processors. The machine-readable storage medium may be non-transitory.
At 1702, the process 1700 includes obtaining a plurality of reference data points. In some embodiments, the plurality of reference data points include data stored in a reference data base, such reference database 116 shown in
At 1704, the process 1700 includes determining a length of a first vector from an origin point to a reference data point of the plurality of reference data points. The reference data point can include the data point 302, the origin can include the origin O, and the first vector can include the reference data point vector 402 shown in
At 1706, the process 1700 includes obtaining an unknown data point associated with content being presented by a display. At 1708, the process 1700 includes determining a length of a second vector from the origin point to the unknown data point. The unknown data point can include the data point 304, the origin can include the origin O, and the second vector can include the unknown data point vector 404 shown in
At 1710, the process 1700 includes projecting a plurality of vectors from the origin point. In one example, the plurality of projected vectors can include the projected vectors shown in
At 1712, the process 1700 includes determining a number of the plurality of vectors between the reference data point and the unknown data point. In some embodiments, determining the number of the plurality of vectors between the reference data point and the unknown data point includes determining whether each vector of the plurality of vectors is to an algebraic right or to an algebraic left of the first vector of the reference data point. The determination of whether a projected vector of the plurality of vectors is to an algebraic right or to an algebraic left of the first vector can include performing a dot product on the projected vector and the first vector. Determining the number of the plurality of vectors between the reference data point and the unknown data point further includes determining whether each vector of the plurality of vectors is to the algebraic right or to the algebraic left of the second vector of the unknown data point. A dot product can also be used. Determining the number of the plurality of vectors between the reference data point and the unknown data point further includes determining the number of the plurality of vectors between the reference data point and the unknown data point. The number of the plurality of vectors includes vectors to the algebraic left of the first vector and to the algebraic right of the second vector or vectors to the algebraic right of the first vector and to the algebraic left of the second vector. One example is shown in
In some examples, the process 1700 further includes storing a first binary value for each vector that is determined to be to the algebraic right of the first vector of the reference data point, and storing a second binary value for each vector that is determined to be to the algebraic left of the first vector of the reference data point. In one example, the first binary value can be a 0 and the second binary value can be a 1. In another example, the first binary value can be a 1 and the second binary value can be a 0.
In some examples, the process 1700 further includes storing a first binary value for each vector that is determined to be to the algebraic right of the second vector of the unknown data point, and storing a second binary value for each vector that is determined to be to the algebraic left of the second vector of the unknown data point. In one example, the first binary value can be a 0 and the second binary value can be a 1. In another example, the first binary value can be a 1 and the second binary value can be a 0.
In some examples, the reference data point is discarded after the length of the first vector of the reference data point is determined and after each vector of the plurality of vectors is determined to be to the algebraic right or to the algebraic left of the first vector. For example, the bits representing the reference data point can be removed from memory. Discarding the reference data point allows much less information to be stored by a matching system.
At 1714, the process 1700 includes estimating an angle between the first vector and the second vector. The angle is estimated using the number of the plurality of vectors. For example, estimating the angle between the first vector of the reference data point and the second vector of the unknown data point includes multiplying a constant by a ratio. The ratio includes the number of the plurality of vectors between the reference data point and the unknown data point divided by a total number of the plurality of vectors. The constant can include a pre-determined value (e.g., 180, 360, or other suitable number).
At 1716, the process 1700 includes determining a distance between the reference data point and the unknown data point. The distance is determined using the estimated angle and the determined lengths of the first vector and the second vector. For example, determining the distance between the reference data point and the unknown data point includes performing a Pythagorean identity calculation using the estimated angle and the determined lengths of the first vector and the second vector.
At 1718, the process 1700 includes identifying one or more candidate data points from the plurality of reference data points. A candidate data point is a candidate for matching the unknown data point. The one or more candidate data points are determined based on determined distances between one or more reference data points and the unknown data point. For example, a nearest neighbor algorithm can be used to determine candidates based on the distances.
In some embodiments, the process 1700 includes determining a match between the unknown data point and a candidate data point. The match is an approximate match based on the candidate data point being a closest data point of the one or more candidate data points to the unknown data point. In some embodiments, the process 1700 can determine the match by comparing the unknown data point with the one or more candidate data points to identify the unknown data point. In some examples, a path pursuit algorithm can be used to identify the unknown data point.
The techniques performed by the systems and methods described herein can be applied to any system that harvests and manipulates large volumes of data. As noted above, illustrative examples of these systems include automated content-based searching systems (e.g., automated content recognition for video-related applications or other suitable application), MapReduce systems, Bigtable systems, pattern recognition systems, facial recognition systems, classification systems, computer vision systems, data compression systems, cluster analysis, or any other suitable system. One of ordinary skill in the art will appreciate that the techniques described herein can be applied to any other system that stores data that is compared to unknown data.
In the context of automated content recognition (ACR), for example, the techniques described above can reduce the amount of data that must be stored in order for a matching system to search and find relationships between unknown and known data groups. For example, Among the many applications of the methods and systems described herein, the vector projection techniques allow identification of media segment of audio and/or video information being presented by a display (e.g., a television (TV), a smart TV, a TV with a cable or satellite feed, an Internet-enabled video set-top box, a mobile device, or any other viewing device). Furthermore, a segment identification system can accurately identify segments of any type whether they are being broadcast, include previously-recorded programming, or include commercial messages. By using the vector projection techniques, a video-based ACR system can reduce the amount of video data that must be stored for reference.
Matching video segments of television programming will be used below as one example of an application of the vector projection techniques described herein. However, one of ordinary skill in the art will appreciate that the techniques and systems described herein can be applied any number of large database searches, analysis, and comparison problems, also known in a general sense as “big data analytics.”
The matching system 1800 can begin a process of matching video segments by first collecting data samples from known video data sources 1818. For example, the video matching server 1804 collects data to build and maintain a reference video database 1816 from a variety of video data sources 1818. The video data sources 1818 can include television programs, movies, or any other suitable video source. The video data sources 1818 can be provided as over-the-air broadcasts, as cable TV channels, as streaming sources from the Internet, and from any other video data source. In some embodiments, the video matching server 1804 can process the received video from the video data source 1818 to generate and collect reference video data points in the reference database 1816, as described with respect to
The video matching server 1804 can store reference video data points for each video program received for a period of time (e.g., a number of days, a number of weeks, a number of months, or any other suitable period of time) in the reference database 1816 until the necessary information is determined. The video matching server 1804 can build and continuously or periodically update the reference database 1816 of television programming samples (e.g., including reference data points, which may also be referred to as cues or cue values). In some examples, the data collected is a compressed representation of the video information sampled from periodic video frames (e.g., every fifth video frame, every tenth video frame, every fifteenth video frame, or other suitable number of frames). In some examples, a number of bytes of data per frame (e.g., 25 bytes, 50 bytes, 75 bytes, 100 bytes, or any other amount of bytes per frame) are collected for each program source. Any number of program sources can be used to obtain video, such as 25 channels, 50 channels, 75 channels, 100 channels, 200 channels, or any other number of program sources. Using the example amount of data, the total data collected during a 24-hour period over three days becomes very large. Therefore, discarding the actual reference video data point bits is advantageous in reducing the storage load of the video matching server 1804.
In one illustrative example,
An example allocation of pixel patches (e.g., pixel patch 1904) is shown in
A mean value (or an average value in some cases) of each pixel patch is taken, and a resulting data record is created and tagged with a time code (or time stamp). For example, a mean value is found for each 10×10 pixel patch array, in which case twenty-four bits of data per twenty-five display buffer locations are produced for a total of 600 bits of pixel information per frame. In one example, a mean of the pixel patch 1904 is calculated, and is shown by pixel patch mean 1908. In one illustrative example, the time code can include an “epoch time,” which representing the total elapsed time (in fractions of a second) since midnight, Jan. 1, 1970. For example, the pixel patch mean 1908 values are assembled with a time code 1912. Epoch time is an accepted convention in computing systems, including, for example, Unix-based systems. Information about the video program, known as metadata, is appended to the data record. The metadata can include any information about a program, such as a program identifier, a program time, a program length, or any other information. The data record including the mean value of a pixel patch, the time code, and metadata, forms a “data point” (also referred to as a “cue”). The data point 1910 is one example of a reference video data point.
A process of identifying unknown video segments begins with steps similar to creating the reference database. For example,
As shown in
The skilled person will know that a reference database 1816 storing actual reference data point bit values creates such a large search space that would require extensive computing resources to search and match data. The vector projection techniques described herein offer a significantly more efficient means to search large databases without the need to actually store large values representing the reference data points (also known as reference data cues).
Process 2100 is illustrated as a logical flow diagram, the operation of which represent a sequence of operations that can be implemented in hardware, computer instructions, or a combination thereof. In the context of computer instructions, the operations represent computer-executable instructions stored on one or more computer-readable storage media that, when executed by one or more processors, perform the recited operations. Generally, computer-executable instructions include routines, programs, objects, components, data structures, and the like that perform particular functions or implement particular data types. The order in which the operations are described is not intended to be construed as a limitation, and any number of the described operations can be combined in any order and/or in parallel to implement the processes.
Additionally, the process 2100 may be performed under the control of one or more computer systems configured with executable instructions and may be implemented as code (e.g., executable instructions, one or more computer programs, or one or more applications) executing collectively on one or more processors, by hardware, or combinations thereof. As noted above, the code may be stored on a machine-readable storage medium, for example, in the form of a computer program comprising a plurality of instructions executable by one or more processors. The machine-readable storage medium may be non-transitory.
At 2102, the process 2100 includes obtaining a plurality of reference video data points. In some embodiments, the plurality of reference video data points include video data extracted from one or more video frames. The reference video data points can include the data point 1910 shown in
At 2104, the process 2100 includes determining a length of a first vector from an origin point to a reference video data point of the plurality of reference video data points. The origin can include the origin O and the first vector can include the reference data point vector 402 shown in
At 2106, the process 2100 includes obtaining an unknown video data point associated with video content being presented by a display. The unknown video data point can include the data point 2010 shown in
At 2108, the process 2100 includes determining a length of a second vector from the origin point to the unknown video data point. The origin can include the origin O and the second vector can include the unknown data point vector 404 shown in
At 2110, the process 2100 includes projecting a plurality of vectors from the origin point. In one example, the plurality of projected vectors can include the projected vectors shown in
At 2112, the process 2100 includes determining a number of the plurality of vectors between the reference video data point and the unknown video data point. In some embodiments, determining the number of the plurality of vectors between the reference video data point and the unknown video data point includes determining whether each vector of the plurality of vectors is to an algebraic right or to an algebraic left of the first vector of the reference video data point. The determination of whether a projected vector of the plurality of vectors is to an algebraic right or to an algebraic left of the first vector can include performing a dot product on the projected vector and the first vector. Determining the number of the plurality of vectors between the reference video data point and the unknown video data point further includes determining whether each vector of the plurality of vectors is to the algebraic right or to the algebraic left of the second vector of the unknown video data point. A dot product can also be used. Determining the number of the plurality of vectors between the reference video data point and the unknown video data point further includes determining the number of the plurality of vectors between the reference video data point and the unknown video data point. The number of the plurality of vectors includes vectors to the algebraic left of the first vector and to the algebraic right of the second vector or vectors to the algebraic right of the first vector and to the algebraic left of the second vector. One example is shown in
In some examples, the process 2100 further includes storing a first binary value for each vector that is determined to be to the algebraic right of the first vector of the reference video data point, and storing a second binary value for each vector that is determined to be to the algebraic left of the first vector of the reference video data point. In one example, the first binary value can be a 0 and the second binary value can be a 1. In another example, the first binary value can be a 1 and the second binary value can be a 0.
In some examples, the process 2100 further includes storing a first binary value for each vector that is determined to be to the algebraic right of the second vector of the unknown video data point, and storing a second binary value for each vector that is determined to be to the algebraic left of the second vector of the unknown video data point. In one example, the first binary value can be a 0 and the second binary value can be a 1. In another example, the first binary value can be a 1 and the second binary value can be a 0.
In some examples, the reference video data point is discarded after the length of the first vector of the reference video data point is determined and after each vector of the plurality of vectors is determined to be to the algebraic right or to the algebraic left of the first vector. For example, the bits representing the reference video data point can be removed from memory. Discarding the reference video data point allows much less information to be stored by a matching system.
At 2114, the process 2100 includes estimating an angle between the first vector and the second vector. The angle is estimated using the number of the plurality of vectors. For example, estimating the angle between the first vector of the reference video data point and the second vector of the unknown video data point includes multiplying a constant by a ratio. The ratio includes the number of the plurality of vectors between the reference video data point and the unknown video data point divided by a total number of the plurality of vectors. The constant can include a pre-determined value (e.g., 180, 360, or other suitable number).
At 2116, the process 2100 includes determining a distance between the reference video data point and the unknown video data point. The distance is determined using the estimated angle and the determined lengths of the first vector and the second vector. For example, determining the distance between the reference video data point and the unknown video data point includes performing a Pythagorean identity calculation using the estimated angle and the determined lengths of the first vector and the second vector.
At 2118, the process 2100 includes identifying one or more candidate video data points from the plurality of reference video data points. A candidate video data point is a candidate for matching the unknown video data point. The one or more candidate video data points are determined based on determined distances between one or more reference video data points and the unknown video data point. For example, a nearest neighbor algorithm can be used to determine candidates based on the distances.
At 2120, the process 2100 includes identifying the video content being presented by the display. The video content being presented by the display is identified by comparing the unknown video data point with the one or more candidate video data points. For example, identifying the video content being presented by the display includes determining a match between the unknown video data point and a candidate video data point. The match is an approximate match based on the candidate video data point being a closest video data point of the one or more candidate video data points to the unknown video data point. In some examples, a path pursuit algorithm can be used to identify the video content being presented.
The nearest neighbor and path pursuit techniques mentioned previously are now described in detail. An example of tracking video transmission using ambiguous cues is given, but the general concept can be applied to any field, such as those described above.
A method for efficient video pursuit is presented. Given a large number of video segments, the system must be able to identify in real time what segment a given query video input is taken from and in what time offset. The segment and offset together are referred to as the location. The method is called video pursuit since it must be able to efficiently detect and adapt to pausing, fast forwarding, rewinding, abrupt switching to other segments and switching to unknown segments. Before being able to pursue live video the data base is processed. Visual cues (a handful of pixel values) are taken from frames every constant fraction of a second and put in specialized data structure (Note that this can also be done in real time). The video pursuit is performed by continuously receiving cues from the input video and updating a set of beliefs or estimates about its current location. Each cue either agrees or disagrees with the estimates, and they are adjusted to reflect the new evidence. A video location is assumed to be the correct one if the confidence in this being true is high enough. By tracking only a small set of possible “suspect” locations, this can be done efficiently.
A method is described for video pursuit but uses mathematical constructs to explain and investigate it. It is the aim of this introduction to give the reader the necessary tools to translate between the two domains. A video signal is comprised of sequential frames. Each can be thought of as a still image. Every frame is a raster of pixels. Each pixel is made out of three intensity values corresponding to the red, green and blue (RGB) make of that pixel's color. In the terminology of this manuscript, a cue is a list of RGB values of a subset of the pixels in a frame and a corresponding time stamp. The number of pixels in a cue is significantly smaller than in a frame, usually between 5 and 15. Being an ordered list of scalar values, the cue values are in fact a vector. This vector is also referred to as a point.
Although these points are in high dimension, usually between 15 and 150, they can be imagined as points in two dimensions. In fact, the illustrations will be given as two dimensional plots. Now, consider the progression of a video and its corresponding cue points. Usually a small change in time produces a small change in pixel values. The pixel point can be viewed as “moving” a little between frames. Following these tiny movements from frame to frame, the cue follows a path in space like a bead would on a bent wire.
In the language of this analogy, in video pursuit the locations of the bead in space (the cue points) are received and the part of wire (path) the bead is following is looked for. This is made significantly harder by two facts. First, the bead does not follow the wire exactly but rather keeps some varying unknown distance from it. Second the wires are all tangled together. These statements are made exact in section 2. The algorithm described below does this in two conceptual steps. When a cue is received, it looks for all points on all the known paths who are sufficiently close to the cue point; these are called suspects. This is done efficiently using the Probabilistic Point Location in Equal Balls algorithm. These suspects are added to a history data structure and the probability of each of them indicating the true location is calculated. This step also includes removing suspect locations who are sufficiently unlikely. This history update process ensures that on the one hand only a small history is kept but on the other hand no probable locations are ever deleted. The generic algorithm is given in Algorithm 1 and illustrated in
The document begins with describing the Probabilistic Point Location in Equal Balls (PPLEB) algorithm in Section 1. It is used in order to perform line 5 in Algorithm 1 efficiently. The ability to perform this search for suspects quickly is crucial for the applicability of this method. Later, in section 2 one possible statistical model is described for performing lines 6 and 7. The described model is a natural choice for the setup. It is also shown how it can be used very efficiently.
Section 1—Probabilistic Point Location in Equal Balls
The following section describes a simple algorithm for performing probabilistic point location in equal balls (PPLEB). In the traditional PLEB (point location in equal balls), one starts with a set of n points x, in 1R d and a specified ball of radius r. The algorithm is given O(poly(n)) preprocessing time to produce an efficient data structure. Then, given a query point x the algorithm is required to return all points x, such that ∥x−xi|≤r. The set of points such that ∥x−∥≤r. geometrically lie within a ball of radius r surrounding the query x (see
The problem of PPLEB and the problem of nearest neighbor search are two similar problems that received much attention in the academic community. In fact, these problems were among the first studied in the field of computational geometry. Many different methods cater to the case where the ambient dimension dis small or constant. These partition the space in different ways and recursively search through the parts. These methods include KD-trees, cover-trees, and others. Although very efficient in low dimension, when the ambient dimension is high, they tend to perform very poorly. This is known as the “curse of dimensionality”. Various approaches attempt to solve this problem while overcoming the curse of dimensionality. The algorithm used herein uses a simpler and faster version of the algorithm and can rely on Local Sensitive Hashing.
Section 1.1 Locality Sensitive Hashing
In the scheme of local sensitive hashing, one devises a family of hash functions H such that:
In words, the probability of x and y being mapped to the same value by h is significantly higher if they are close to each other.
For the sake of clarity, let us first deal with a simplified scenario where all incoming vectors are of the same length r′ and r′>√{square root over (2r)}. The reason for the latter condition will become clear later. First a random function uϵU is defined, which separates between x and y according to the angle between them. Let {right arrow over (u)} be a random vector chosen uniformly from the unit sphere Sd−1 and let u(x)=sign ({right arrow over (u)}·x) (See
The family of functions H is set to be a cross product oft independent copies of u, i.e. h(x)=[u1(x), . . . , ut(x)]. Intuitively, one would like to have that if h(x)=h(y) then x and y are likely to be close to each other. Let us quantify that. First, compute the expected number of false positive mistakes nfp. These are the cases for which h(x)=h(y) but ∥x−y∥>2r. A value t is found for which nfp is no more than 1, i.e. one is not expected to be wrong.
E[nft]≤n(1−2p)t≤1→t≥log(1/n)/log(1−2p)
Now, the probability that h(x)=h(y) given that they are neighbors is computed:
Note here that one must have that 2p<1 which requires r′>√{square root over (2r)}. This might not sound like a very high success probability. Indeed, 1/√{square root over (n)} is significantly smaller than ½. The next section will describe how to boost this probability up to ½.
Section 1.2 The Point Search Algorithm
Each function h maps every point in space to a bucket. Define the bucket function Bh:d→2[n] of a point x with respect to hash function h as Bh(x)∝{xi|h(xi)=h(x)}. The data structure maintained is m=O(√{square root over (n)}) instances of bucket functions [Bh1, . . . , Bhm]. When one searches for a point x, the function returns B(x)=∪i Bh
Pr(xi∈B(x)|∥xi−x∥≤r)≥½
E[|B(x)∩{xi|∥x−xi∥>2r}|/]≤√{square root over (n)}.
In other words, while with probability at least ½ each neighbor of x is found, one is not likely to find many non-neighbors.
Section 1.3 Dealing with Different Radii Input Vectors
The previous sections only dealt with searching through vectors of the same length, namely r′. Now described is how one can use the construction as a building block to support a search in different radii. As seen in
Section 2 the Path Pursuit Problem
In the path pursuit problem, a fixed path in space is given along with the positions of a particle in a sequence of time points. The terms particle, cue, and point will be used interchangeably. The algorithm is required to output the position of the particle on the path. This is made harder by a few factors: The particle only follows the path approximately; The path can be discontinuous and intersect itself many times; Both particle and path positions are given in a sequence of time points (different for each).
It is important to note that this problem can simulate tracking a particle on any number of paths. This is simply done by concatenating the paths into one long path and interpreting the resulting position as the position on the individual paths.
More precisely, let path P be parametric curve P:→
d. The curve parameter will be referred to as the time. The points on the path that are known to us are given in arbitrary time points ti, i.e. n pairs (ti, P(ti)) are given. The particle follows the path but its positions are given in different time points, as shown in
Section 2.1 Likelihood Estimation
Since the particle does not follow the path exactly and since the path can intersect itself many times it is usually impossible to positively identify the position on the path the particle is actually on. Therefore, a probability distribution is computed on all possible path locations. If a location probability is significantly probable, the particle position is assumed to be known. The following section describes how this can be done efficiently.
If the particle is following the path, then the time difference between the particle time stamp and the offset of the corresponding points on P should be relatively fixed. In other words, if x(t′) is currently in offset t on the path then it should be close to P(t). Also, τ seconds ago it should have been in offset t−τ. Thus x(t′−τ) should be close to P(t−τ) (note that if the particle is intersecting the path, and x(t′) is close to P(t) temporarily, it is unlikely that x(t′−τ) and P(t−τ) will also be close). Define the relative offset as Δ=t−t′. Notice that as long as the particle is following the path the relative offset Δ remains unchanged. Namely, x(t′) is close to P(t′+Δ).
The maximum likelihood relative offset is obtained by calculating:
In words, the most likely relative offset is the one for which the history of the particle is most likely. This equation however cannot be solved without a statistical model. This model must quantify: How tightly x follows the path; How likely it is that x′)umps” between locations; How smooth the path and particle curves are between the measured points.
Section 2.2 Time Discounted Binning
Now described is a statistical model for estimating the likelihood function. The model makes the assumption that the particle's deviation away from the path distributes normally with standard deviation ar. It also assumes that at any given point in time, there is some non-zero probability the particle will abruptly switch to another path. This is manifested by an exponential discount with time for past points. Apart for being a reasonable choice for a modeling point of view this model also has the advantage of being efficiently updateable. For some constant time unit 1, set the likelihood function to be proportional to ƒ which is defined as follows:
Here α<<1 is a scale coefficient and ζ>0 is the probability that the particle will jump to a random location on the path in a given time unit.
Updating the function ƒ efficiently can be achieved using the following simple observation.
Moreover, since α<<1 if ∥x(t′m)−P(ti)∥≥r, the follow occurs:
This is an important property of the likelihood function since the sum update can now performed only over the neighbors of x(t′j) and not the entire path. Denote by S the set of (ti, P(ti)) such that ∥x(t′m)−P(ti)∥≤r. The follow equation occurs:
This is described in Algorithm 2.2 below. The term ƒ is used as a sparse vector that receives also negative integer indices. The set S is the set of all neighbors of x(ti) on the path and can be computed quickly using the PPLEB algorithm. It is easy to verify that if the number of neighbors of x(ti) is bounded by some constant nnear then the number of non-zeros in the vector ƒ is bounded by nnear/ζ which is only a constant factor larger. The final stage of the algorithm is to output a specific value of δ if (└δ/τ┘) is above some threshold value.
In
Substantial variations may be made in accordance with specific requirements. For example, customized hardware might also be used, and/or particular elements might be implemented in hardware, software (including portable software, such as applets, etc.), or both. Further, connection to other access or computing devices such as network input/output devices may be employed.
In the foregoing specification, aspects of the invention are described with reference to specific embodiments thereof, but those skilled in the art will recognize that the invention is not limited thereto. Various features and aspects of the above-described invention may be used individually or jointly. Further, embodiments can be utilized in any number of environments and applications beyond those described herein without departing from the broader spirit and scope of the specification. The specification and drawings are, accordingly, to be regarded as illustrative rather than restrictive.
In the foregoing description, for the purposes of illustration, methods were described in a particular order. It should be appreciated that in alternate embodiments, the methods may be performed in a different order than that described. It should also be appreciated that the methods described above may be performed by hardware components or may be embodied in sequences of machine-executable instructions, which may be used to cause a machine, such as a general-purpose or special-purpose processor or logic circuits programmed with the instructions to perform the methods. These machine-executable instructions may be stored on one or more machine readable mediums, such as CD-ROMs or other type of optical disks, floppy diskettes, ROMs, RAMs, EPROMs, EEPROMs, magnetic or optical cards, flash memory, or other types of machine-readable mediums suitable for storing electronic instructions. Alternatively, the methods may be performed by a combination of hardware and software.
Where components are described as being configured to perform certain operations, such configuration can be accomplished, for example, by designing electronic circuits or other hardware to perform the operation, by programming programmable electronic circuits (e.g., microprocessors, or other suitable electronic circuits) to perform the operation, or any combination thereof.
While illustrative embodiments of the application have been described in detail herein, it is to be understood that the inventive concepts may be otherwise variously embodied and employed, and that the appended claims are intended to be construed to include such variations, except as limited by the prior art.
This application claims the benefit of U.S. Provisional Application No. 62/149,193, filed Apr. 17, 2015, which is hereby incorporated by reference in its entirety. This application is related to U.S. patent application Ser. No. 12/788,748, filed May 27, 2010, U.S. patent application Ser. No. 12/788,721, filed May 27, 2010, U.S. patent application Ser. No. 14/089,003, filed Nov. 25, 2013, U.S. patent application Ser. No. 14/217,075, filed Mar. 17, 2014, U.S. patent application Ser. No. 14/217,039, filed Mar. 17, 2014, U.S. patent application Ser. No. 14/217,094, filed Mar. 17, 2014, U.S. patent application Ser. No. 14/217,375, filed Mar. 17, 2014, U.S. patent application Ser. No. 14/217,425, filed Mar. 17, 2014, U.S. patent application Ser. No. 14/217,435, filed Mar. 17, 2014, U.S. patent application Ser. No. 14/551,933, filed Nov. 24, 2014, and U.S. patent application Ser. No. 14/763,158, filed Dec. 23, 2014, all of which are hereby incorporated by reference in their entirety.
| Number | Name | Date | Kind |
|---|---|---|---|
| 4677466 | Lert, Jr. et al. | Jun 1987 | A |
| 4697209 | Kiewit et al. | Sep 1987 | A |
| 4739398 | Thomas et al. | Apr 1988 | A |
| 5019899 | Boles et al. | May 1991 | A |
| 5319453 | Copriviza et al. | Jun 1994 | A |
| 5436653 | Ellis et al. | Jul 1995 | A |
| 5481294 | Thomas et al. | Jan 1996 | A |
| 5557334 | Legate | Sep 1996 | A |
| 5572246 | Ellis et al. | Nov 1996 | A |
| 5812286 | Li | Sep 1998 | A |
| 5826165 | Echeita et al. | Oct 1998 | A |
| 5918223 | Blum et al. | Jun 1999 | A |
| 6008802 | Goldschmidt et al. | Dec 1999 | A |
| 6035177 | Moses et al. | Mar 2000 | A |
| 6064764 | Bhaskaran et al. | May 2000 | A |
| 6381362 | Deshpande et al. | Apr 2002 | B1 |
| 6415438 | Blackketter et al. | Jul 2002 | B1 |
| 6463585 | Hendricks et al. | Oct 2002 | B1 |
| 6469749 | Dimitrova | Oct 2002 | B1 |
| 6577346 | Perlman | Jun 2003 | B1 |
| 6577405 | Kranz et al. | Jun 2003 | B2 |
| 6628801 | Powell et al. | Sep 2003 | B2 |
| 6647548 | Lu et al. | Nov 2003 | B1 |
| 6675174 | Bolle et al. | Jan 2004 | B1 |
| 6771316 | Iggulden | Aug 2004 | B1 |
| 6804659 | Graham et al. | Oct 2004 | B1 |
| 6978470 | Swix et al. | Dec 2005 | B2 |
| 6990453 | Wang et al. | Jan 2006 | B2 |
| 7028327 | Dougherty et al. | Apr 2006 | B1 |
| 7039930 | Goodman et al. | May 2006 | B1 |
| 7050068 | Bastos et al. | May 2006 | B1 |
| 7051351 | Goldman et al. | May 2006 | B2 |
| 7064796 | Roy et al. | Jun 2006 | B2 |
| 7089575 | Agnihotri et al. | Aug 2006 | B2 |
| 7098959 | Mishima | Aug 2006 | B2 |
| 7136875 | Anderson et al. | Nov 2006 | B2 |
| 7210157 | Devara | Apr 2007 | B2 |
| 7346512 | Wang et al. | Mar 2008 | B2 |
| 7421723 | Harkness et al. | Sep 2008 | B2 |
| 7590998 | Hanley | Sep 2009 | B2 |
| 7623823 | Zito et al. | Nov 2009 | B2 |
| 7787696 | Wilhelm | Aug 2010 | B2 |
| 7793318 | Deng | Sep 2010 | B2 |
| 7933451 | Kloer | Apr 2011 | B2 |
| 8001571 | Schwartz et al. | Aug 2011 | B1 |
| 8094872 | Yagnik et al. | Jan 2012 | B1 |
| 8171004 | Kaminski, Jr. et al. | May 2012 | B1 |
| 8171030 | Peira et al. | May 2012 | B2 |
| 8175413 | Ioffe et al. | May 2012 | B1 |
| 8189945 | Stojancic et al. | May 2012 | B2 |
| 8195689 | Ramanathan et al. | Jun 2012 | B2 |
| 8229227 | Stojancic et al. | Jul 2012 | B2 |
| 8335786 | Peira et al. | Dec 2012 | B2 |
| 8364703 | Ramanathan et al. | Jan 2013 | B2 |
| 8385644 | Stojancic et al. | Feb 2013 | B2 |
| 8392789 | Biscondi et al. | Mar 2013 | B2 |
| 8494234 | Djordjevic et al. | Jul 2013 | B1 |
| 8522283 | Laligand et al. | Aug 2013 | B2 |
| 8595781 | Neumeier et al. | Nov 2013 | B2 |
| 8625902 | Baheti et al. | Jan 2014 | B2 |
| 8769854 | Battaglia | Jul 2014 | B1 |
| 8776105 | Sinha et al. | Jul 2014 | B2 |
| 8832723 | Sinha et al. | Sep 2014 | B2 |
| 8856817 | Sinha et al. | Oct 2014 | B2 |
| 8893167 | Sinha et al. | Nov 2014 | B2 |
| 8893168 | Sinha et al. | Nov 2014 | B2 |
| 8898714 | Neumeier et al. | Nov 2014 | B2 |
| 8918804 | Sinha et al. | Dec 2014 | B2 |
| 8918832 | Sinha et al. | Dec 2014 | B2 |
| 8930980 | Neumeier et al. | Jan 2015 | B2 |
| 8959202 | Haitsma et al. | Feb 2015 | B2 |
| 9055309 | Neumeier et al. | Jun 2015 | B2 |
| 9055335 | Neumeier et al. | Jun 2015 | B2 |
| 9071868 | Neumeier et al. | Jun 2015 | B2 |
| 9094714 | Neumeier et al. | Jul 2015 | B2 |
| 9094715 | Neumeier et al. | Jul 2015 | B2 |
| 9449090 | Neumeier et al. | Sep 2016 | B2 |
| 9465867 | Hoarty | Oct 2016 | B2 |
| 20010039658 | Walton | Nov 2001 | A1 |
| 20010044992 | Jahrling | Nov 2001 | A1 |
| 20020026635 | Wheeler et al. | Feb 2002 | A1 |
| 20020054695 | Bjorn et al. | May 2002 | A1 |
| 20020056088 | Silva, Jr. et al. | May 2002 | A1 |
| 20020059633 | Harkness et al. | May 2002 | A1 |
| 20020100041 | Rosenberg et al. | Jul 2002 | A1 |
| 20020105907 | Bruekers et al. | Aug 2002 | A1 |
| 20020120925 | Logan | Aug 2002 | A1 |
| 20020122042 | Bates | Sep 2002 | A1 |
| 20020162117 | Pearson et al. | Oct 2002 | A1 |
| 20020162118 | Levy et al. | Oct 2002 | A1 |
| 20030026422 | Gerheim et al. | Feb 2003 | A1 |
| 20030086341 | Wells | May 2003 | A1 |
| 20030121037 | Swix et al. | Jun 2003 | A1 |
| 20030121046 | Roy et al. | Jun 2003 | A1 |
| 20030147561 | Faibish et al. | Aug 2003 | A1 |
| 20030188321 | Shoff et al. | Oct 2003 | A1 |
| 20040045020 | Witt et al. | Mar 2004 | A1 |
| 20040059708 | Dean et al. | Mar 2004 | A1 |
| 20040216171 | Barone et al. | Oct 2004 | A1 |
| 20040221237 | Foote et al. | Nov 2004 | A1 |
| 20040226035 | Hauser | Nov 2004 | A1 |
| 20040240562 | Bargeron et al. | Dec 2004 | A1 |
| 20050015795 | Iggulden | Jan 2005 | A1 |
| 20050015796 | Bruckner et al. | Jan 2005 | A1 |
| 20050027766 | Ben | Feb 2005 | A1 |
| 20050066352 | Herley | Mar 2005 | A1 |
| 20050172312 | Lienhart et al. | Aug 2005 | A1 |
| 20050207416 | Rajkotia | Sep 2005 | A1 |
| 20050209065 | Schlosser et al. | Sep 2005 | A1 |
| 20050235318 | Grauch et al. | Oct 2005 | A1 |
| 20060029368 | Harville | Feb 2006 | A1 |
| 20060031914 | Dakss et al. | Feb 2006 | A1 |
| 20060133647 | Werner et al. | Jun 2006 | A1 |
| 20060153296 | Deng | Jul 2006 | A1 |
| 20060155952 | Haas | Jul 2006 | A1 |
| 20060173831 | Basso et al. | Aug 2006 | A1 |
| 20060187358 | Lienhart et al. | Aug 2006 | A1 |
| 20060195857 | Wheeler et al. | Aug 2006 | A1 |
| 20060195860 | Eldering et al. | Aug 2006 | A1 |
| 20060245724 | Hwang et al. | Nov 2006 | A1 |
| 20060245725 | Lim | Nov 2006 | A1 |
| 20060253330 | Maggio et al. | Nov 2006 | A1 |
| 20070033608 | Eigeldinger | Feb 2007 | A1 |
| 20070050832 | Wright et al. | Mar 2007 | A1 |
| 20070061724 | Slothouber et al. | Mar 2007 | A1 |
| 20070061831 | Savoor et al. | Mar 2007 | A1 |
| 20070094696 | Sakai | Apr 2007 | A1 |
| 20070109449 | Cheung | May 2007 | A1 |
| 20070113263 | Chatani | May 2007 | A1 |
| 20070139563 | Zhong | Jun 2007 | A1 |
| 20070143796 | Malik | Jun 2007 | A1 |
| 20070168409 | Cheung | Jul 2007 | A1 |
| 20070180459 | Smithpeters et al. | Aug 2007 | A1 |
| 20070192782 | Ramaswamy | Aug 2007 | A1 |
| 20070250901 | McIntire et al. | Oct 2007 | A1 |
| 20070261070 | Brown et al. | Nov 2007 | A1 |
| 20070261075 | Glasberg | Nov 2007 | A1 |
| 20070271300 | Ramaswamy | Nov 2007 | A1 |
| 20070274537 | Srinivasan | Nov 2007 | A1 |
| 20070300280 | Turner et al. | Dec 2007 | A1 |
| 20080044102 | Ekin | Feb 2008 | A1 |
| 20080046945 | Hanley | Feb 2008 | A1 |
| 20080089551 | Heather et al. | Apr 2008 | A1 |
| 20080138030 | Bryan et al. | Jun 2008 | A1 |
| 20080155588 | Roberts et al. | Jun 2008 | A1 |
| 20080155627 | O'Connor et al. | Jun 2008 | A1 |
| 20080172690 | Kanojia et al. | Jul 2008 | A1 |
| 20080208891 | Wang et al. | Aug 2008 | A1 |
| 20080240562 | Fukuda et al. | Oct 2008 | A1 |
| 20080263620 | Berkvens et al. | Oct 2008 | A1 |
| 20080276266 | Huchital et al. | Nov 2008 | A1 |
| 20080310731 | Stojancic et al. | Dec 2008 | A1 |
| 20080313140 | Pereira et al. | Dec 2008 | A1 |
| 20090028517 | Shen et al. | Jan 2009 | A1 |
| 20090052784 | Covell et al. | Feb 2009 | A1 |
| 20090087027 | Eaton et al. | Apr 2009 | A1 |
| 20090088878 | Otsuka et al. | Apr 2009 | A1 |
| 20090100361 | Abello et al. | Apr 2009 | A1 |
| 20090131861 | Braig et al. | May 2009 | A1 |
| 20090172728 | Shkedi et al. | Jul 2009 | A1 |
| 20090172746 | Aldrey et al. | Jul 2009 | A1 |
| 20090213270 | Ismert et al. | Aug 2009 | A1 |
| 20090235312 | Morad et al. | Sep 2009 | A1 |
| 20100083299 | Nelson | Apr 2010 | A1 |
| 20100115543 | Falcon | May 2010 | A1 |
| 20100125870 | Ukawa et al. | May 2010 | A1 |
| 20100166257 | Wredenhagen | Jul 2010 | A1 |
| 20100199295 | Katpelly et al. | Aug 2010 | A1 |
| 20100235486 | White et al. | Sep 2010 | A1 |
| 20100269138 | Krikorian et al. | Oct 2010 | A1 |
| 20100306805 | Neumeier et al. | Dec 2010 | A1 |
| 20100306808 | Neumeier et al. | Dec 2010 | A1 |
| 20110015996 | Kassoway et al. | Jan 2011 | A1 |
| 20110026761 | Radhakrishnan et al. | Feb 2011 | A1 |
| 20110041154 | Olson | Feb 2011 | A1 |
| 20110055552 | Francis et al. | Mar 2011 | A1 |
| 20110096955 | Voloshynovskiy et al. | Apr 2011 | A1 |
| 20110251987 | Buchheit | Apr 2011 | A1 |
| 20110247042 | Mallinson | Oct 2011 | A1 |
| 20110289099 | Quan | Nov 2011 | A1 |
| 20110299770 | Vaddadi et al. | Dec 2011 | A1 |
| 20120017240 | Shkedi | Jan 2012 | A1 |
| 20120054143 | Doig et al. | Mar 2012 | A1 |
| 20120095958 | Pereira | Apr 2012 | A1 |
| 20120158511 | Lucero et al. | Jun 2012 | A1 |
| 20120174155 | Mowrey et al. | Jul 2012 | A1 |
| 20120177249 | Levy et al. | Jul 2012 | A1 |
| 20120185566 | Nagasaka | Jul 2012 | A1 |
| 20120272259 | Cortes | Oct 2012 | A1 |
| 20120317240 | Wang | Dec 2012 | A1 |
| 20120324494 | Burger et al. | Dec 2012 | A1 |
| 20130007191 | Klappert et al. | Jan 2013 | A1 |
| 20130042262 | Riethmueller | Feb 2013 | A1 |
| 20130054356 | Richman et al. | Feb 2013 | A1 |
| 20130067523 | Etsuko et al. | Mar 2013 | A1 |
| 20130070847 | Iwamoto et al. | Mar 2013 | A1 |
| 20130139209 | Urrabazo et al. | May 2013 | A1 |
| 20130202150 | Sinha et al. | Aug 2013 | A1 |
| 20130209065 | Yeung | Aug 2013 | A1 |
| 20130290502 | Bilobrov | Oct 2013 | A1 |
| 20130297727 | Levy | Nov 2013 | A1 |
| 20130318096 | Cheung | Nov 2013 | A1 |
| 20140016696 | Nelson | Jan 2014 | A1 |
| 20140082663 | Neumeier et al. | Mar 2014 | A1 |
| 20140088742 | Srinivasan | Mar 2014 | A1 |
| 20140130092 | Kunisetty | May 2014 | A1 |
| 20140188487 | Perez Gonzalez | Jul 2014 | A1 |
| 20140193027 | Scherf | Jul 2014 | A1 |
| 20140195548 | Harron | Jul 2014 | A1 |
| 20140201769 | Neumeier et al. | Jul 2014 | A1 |
| 20140201772 | Neumeier et al. | Jul 2014 | A1 |
| 20140219554 | Yamaguchi et al. | Aug 2014 | A1 |
| 20140237576 | Zhang | Aug 2014 | A1 |
| 20140258375 | Munoz | Sep 2014 | A1 |
| 20140270489 | Jaewhan et al. | Sep 2014 | A1 |
| 20140270504 | Baum et al. | Sep 2014 | A1 |
| 20140270505 | McCarthy | Sep 2014 | A1 |
| 20140282671 | McMillan | Sep 2014 | A1 |
| 20140344880 | Geller et al. | Nov 2014 | A1 |
| 20150100979 | Moskowitz et al. | Apr 2015 | A1 |
| 20150112988 | Pereira et al. | Apr 2015 | A1 |
| 20150163545 | Freed et al. | Jun 2015 | A1 |
| 20150181311 | Navin et al. | Jun 2015 | A1 |
| 20150382075 | Neumeier et al. | Dec 2015 | A1 |
| 20160227261 | Neumeier et al. | Aug 2016 | A1 |
| 20160307043 | Neumeier | Oct 2016 | A1 |
| 20170017645 | Neumeier et al. | Jan 2017 | A1 |
| 20170017651 | Neumeier et al. | Jan 2017 | A1 |
| 20170017652 | Neumeier et al. | Jan 2017 | A1 |
| 20170019716 | Neumeier et al. | Jan 2017 | A1 |
| 20170019719 | Neumeier et al. | Jan 2017 | A1 |
| 20170026671 | Neumeier et al. | Jan 2017 | A1 |
| 20170032033 | Neumeier et al. | Feb 2017 | A1 |
| 20170032034 | Neumeier et al. | Feb 2017 | A1 |
| 20170134770 | Neumeier et al. | May 2017 | A9 |
| Number | Date | Country |
|---|---|---|
| 2501316 | Sep 2005 | CA |
| 1557096 | Dec 2004 | CN |
| 101162470 | Apr 2008 | CN |
| 1681304 | Jul 2010 | CN |
| 102377960 | Mar 2012 | CN |
| 248 533 | Aug 1994 | EP |
| 1578126 | Sep 2005 | EP |
| 2 084 624 | Aug 2009 | EP |
| 2 352 289 | Aug 2011 | EP |
| 2 541 963 | Jan 2013 | EP |
| 2457694 | Aug 2009 | GB |
| 0144992 | Jun 2001 | WO |
| 2005101998 | Nov 2005 | WO |
| 2007114796 | Oct 2007 | WO |
| 2008065340 | Jun 2008 | WO |
| 2009131861 | Oct 2009 | WO |
| 2009150425 | Dec 2009 | WO |
| 2010135082 | Nov 2010 | WO |
| 2011090540 | Jul 2011 | WO |
| 2012057724 | May 2012 | WO |
| 2012108975 | Aug 2012 | WO |
| 2012170451 | Dec 2012 | WO |
| 2014142758 | Sep 2014 | WO |
| 2014145929 | Sep 2014 | WO |
| 2015100372 | Jul 2015 | WO |
| 2016123495 | Aug 2016 | WO |
| 2016168556 | Oct 2016 | WO |
| 2017011758 | Jan 2017 | WO |
| 2017011792 | Jan 2017 | WO |
| Entry |
|---|
| International Search Report and Written Opinion dated Mar. 31, 2015 for PCT Application No. PCT/US2014/072255, 8 pages. |
| International Search Report and Written Opinion dated Apr. 26, 2016 for PCT Application No. PCT/US2016/015681,13 pages. |
| “How to: Watch from the beginning |About Dish” (Dec. 31, 2014) XP055265764, retrieved on Apr. 15, 2016 from URL:http://about.dish.com/blog/hopper/how-watch-beginning 2 pages. |
| International Search Report and Written Opinion dated Oct. 12, 2016 for PCT Application No. PCT/US2016/042522,13 pages. |
| International Search Report and Written Opinion dated Oct. 11, 2016 for PCT Application No. PCT/US2016/042621, 13 pages. |
| International Search Report and Written Opinion dated Oct. 20, 2016 for PCT Application No. PCT/US2016/042611,12 pages. |
| Scouarnec et al., “Cache policies for cloud-based systems:To keep or not to keep”, 2014 IEEE 7th International Conference on Cloud Computing, IEEE XP032696624, Jun. 27, 2014, pp. 1-8. |
| International Search Report and Written Opinion dated Oct. 25, 2016 for PCT Application No. PCT/US2016/042564, 14 pages. |
| Anonymous; “Cache (computing)” Wikipedia, the free encyclopedia, URL:http://en.wikipedia.org/w/index.phpti tle=Cache (computing)&oldid=474222804, Jan. 31, 2012; 6 pages. |
| International Search Report and Written Opinion dated Oct. 24, 2016 for PCT Application No. PCT/US2016/042557, 11 pages. |
| Anil K. Jain, “Image Coding via a Nearest Neighbors Image Model” IEEE Transactions on Communications, vol. Com-23, No. 3, Mar. 1975, pp. 318-331. |
| Lee et al., “Fast Video Search Algorithm for Large Video Database Using Adjacent Pixel Intensity Difference Quantization Histogram Feature” International Journal of Computer Science and Network Security, vol. 9, No. 9, Sep. 2009, pp. 214-220. |
| Li et al., A Confidence Based Recognition System for TV Commercial Extraction, Conferences in Research and Practice in Information Technology vol. 75, 2008. |
| International Search Report and Written Opinion dated Jul. 27, 2011 for PCT Application No. PCT/US2010/057153, 8 pages. |
| International Search Report and Written Opinion dated Aug. 31, 2011 for PCT Application No. PCT/US2010/057155, 8 pages. |
| International Search Report and Written Opinion dated Aug. 26, 2014 for PCT Application No. PCT/US2014/030782; 11 pages. |
| International Search Report and Written Opinion dated Jul. 21, 2014 for PCT Application No. PCT/US2014/030795; 10 pages. |
| International Search Report and Written Opinion, dated Jul. 25, 2014 for PCT Application No. PCT/US2014/030805, 10 pages. |
| Extended European Search Report dated Mar. 7, 2013 for European Application No. 12178359.1, 8 pages. |
| Extended European Search Report dated Oct. 11, 2013 for European Application No. 10844152.8, 19 pages. |
| Kabal (P.), Ramachandran (R.P.): The computation of line spectral frequencies using Chebyshev polynomials, IEEE Trans. on ASSP, vol. 34, No. 6, pp. 1419-1426, 1986. |
| Itakura (F.): Line spectral representation of linear predictive coefficients of speech signals, J. Acoust. Soc. Amer., vol. 57, Supplement No. 1, S35, 1975, 3 pages. |
| Bistritz (Y.), Pellerm (S.): Immittance Spectral Pairs (ISP) for speech encoding, Proc. ICASSP'93, pp. 11-9 to 11-12. |
| International Search Report and Written Opinion dated Mar. 8, 2016 for PCT Application No. PCT/ US2015/062945; 9 pages. |
| Extended European Search Report dated Dec. 21, 2016 for European Application No. 14763506.4, 11 pages. |
| Extended European Search Report dated Nov. 23, 2016 for European Application No. 14764182.3, 10 pages. |
| Extended European Search Report dated Jan. 24, 2017 for European Application No. 14762850.7, 12 pages. |
| Extended European Search Report dated Jun. 16, 2017, for European Patent Application No. 14873564.0, 8 pages. |
| U.S. Appl. No. 14/551,933 , “Final Office Action”, dated May 23, 2016, 19 pages. |
| U.S. Appl. No. 14/551,933 , “Non-Final Office Action”, dated Oct. 17, 2016, 15 pages. |
| U.S. Appl. No. 14/551,933 , “Non-Final Office Action”, dated Dec. 31, 2015, 24 pages. |
| U.S. Appl. No. 14/551,933 , “Notice of Allowance”, dated Mar. 21, 2017, 8 pages. |
| U.S. Appl. No. 14/217,039 , “Non-Final Office Action”, dated May 23, 2014, 27 pages. |
| U.S. Appl. No. 14/217,039 , “Final Office Action”, dated Nov. 7, 2014, 15 pages. |
| U.S. Appl. No. 14/217,039 , “Notice of Allowance”, dated Jan. 29, 2015, 8 pages. |
| U.S. Appl. No. 14/678,856 , “Non-Final Office Action”, dated Dec. 1, 2015, 28 pages. |
| U.S. Appl. No. 14/678,856 , “Notice of Allowance”, dated May 20, 2016, 9 pages. |
| U.S. Appl. No. 14/217,075, “Non-Final Office Action”, dated Jul. 16, 2014, 39 pages. |
| U.S. Appl. No. 14/217,075, “Notice of Allowance ”, dated Feb. 20, 2015, 51 pages. |
| U.S. Appl. No. 14/217,094, “Notice of Allowance ”, dated Sep. 4, 2014, 30 pages. |
| U.S. Appl. No. 14/217,375, “Non-Final Office Action”, dated Apr. 1, 2015, 39 pages. |
| U.S. Appl. No. 14/217,375, “Notice of Allowance”, dated Apr. 1, 2015, 31 pages. |
| U.S. Appl. No. 14/217,425, “Non-Final Office Action”, dated Apr. 7, 2015, 12 pages. |
| U.S. Appl. No. 14/217,425, “Notice of Allowance”, dated May 20, 2015, 15 pages. |
| U.S. Appl. No. 14/217,435, “Non-Final Office Action”, dated Nov. 24, 2014, 9 pages. |
| U.S. Appl. No. 14/217,435, “Notice of Allowance”, dated Jun. 5, 2015, 9 pages. |
| U.S. Appl. No. 15/011,099 , “First Action Interview Office Action Summary”, dated May 9, 2017, 6 pages. |
| U.S. Appl. No. 15/011,099 , “First Action Interview Pilot Program Pre-Interview Communication”, dated Feb. 28, 2017, 5 pages. |
| U.S. Appl. No. 12/788,721 , “Non-Final Office Action”, dated Mar. 28, 2012, 15 Pages. |
| U.S. Appl. No. 12/788,721 , “Final Office Action”, dated Aug. 15, 2012, 22 Pages. |
| U.S. Appl. No. 12/788,721 , “Notice of Allowance”, dated Aug. 15, 2013, 16 Pages. |
| U.S. Appl. No. 14/763,158 , “Non-Final Office Action”, dated Jun. 27, 2016, 16 Pages. |
| U.S. Appl. No. 14/763,158 , “Final Office Action”, dated Sep. 7, 2016, 12 Pages. |
| U.S. Appl. No. 14/763,158 , “Notice of Allowance”, dated Mar. 17, 2016, 8 Pages. |
| U.S. Appl. No. 14/807,849 , “Non-Final Office Action”, dated Nov. 25, 2015, 12 Pages. |
| U.S. Appl. No. 14/807,849 , “Final Office Action”, dated Apr. 19, 2016, 13 pages. |
| U.S. Appl. No. 14/807,849 , “Non-Final Office Action”, dated Feb. 28, 2017, 10 Pages. |
| U.S. Appl. No. 14/089,003 , “Notice of Allowance”, dated Jul. 30, 2014, 24 Pages. |
| U.S. Appl. No. 12/788,748 , “Non-Final Office Action”, dated Jan. 10, 2013, 10 Pages. |
| U.S. Appl. No. 12/788,748 , “Final Office Action”, dated Nov. 21, 2013, 13 Pages. |
| U.S. Appl. No. 12/788,748 , “Notice of Allowance”, dated Mar. 6, 2014, 7 Pages. |
| U.S. Appl. No. 14/953,994 , “Non-Final Office Action”, dated Mar. 3, 2016, 34 Pages. |
| U.S. Appl. No. 14/953,994 , “Final Office Action”, dated Jun. 1, 2016, 36 Pages. |
| U.S. Appl. No. 14/953,994 , “Notice of Allowance”, dated Aug. 31, 2016, 15 Pages. |
| U.S. Appl. No. 14/807,849 , “Final Office Action”, dated Jun. 22, 2017, 10 pages. |
| U.S. Appl. No. 15/011,099 , “Final Office Action”, dated Jul. 24, 2017, 22 pages. |
| U.S. Appl. No. 15/240,801 , “Non-Final Office Action”, dated Aug. 11, 2017, 18 pages. |
| U.S. Appl. No. 15/240,815 , “Non-Final Office Action”, dated Aug. 23, 2017, 15 pages. |
| International Search Report and Written Opinion issued in International Application No. PCT/US2016/027691 dated Jun. 24, 2016, 13 pages. |
| Gionis et al., “Similarity Search in High Dimension via Hashing”, Proceedings of the 25th VLDB Conference, 1999, 12 pages. |
| Huang , “Bounded Coordinate System Indexing for Real-time Video Clip Search”, Retrieved from the Internet: URL:http://staff.itee.uq.edu.aujzxf/_papers/TOIS.pdf, Jan. 1, 2009, 32 pages. |
| Kim et al., “Edge-Based Spatial Descriptor Using Color Vector Angle for Effective Image Retrieval”, Modeling Decisions for Artificial Intelligence; [Lecture Notes in Computer Science; Lecture Notes in Artificial Intelligence, Jul. 1, 2005, pp. 365-375. |
| Liu et al., “Near-duplicate video retrieval”, ACM Computing Surveys, vol. 45, No. 4, Aug. 30, 2013, pp. 1-23. |
| Number | Date | Country | |
|---|---|---|---|
| 20160307043 A1 | Oct 2016 | US |
| Number | Date | Country | |
|---|---|---|---|
| 62149193 | Apr 2015 | US |
