System and method for determining common patterns in multimedia content elements based on key points

Abstract

A system and method for method for determining common patterns based on key points in multimedia data elements (MMDEs). The method includes: identifying a plurality of candidate key points in each of the plurality of MMDEs, wherein a size of each candidate key point is equal to a predetermined size and a scale of each candidate key point is equal to a predetermined scale; analyzing the identified candidate key points to determine a set of properties for each candidate key point; comparing the sets of properties of the plurality of candidate key points of each MMDE; selecting, for each MMDE, a plurality of key points from among the candidate key points based on the comparison; generating, based on the key points for each MMDE, a signature for the MMDE; and comparing the signatures of the plurality of MMDEs to output at least one common pattern among the plurality of MMDEs.

Description

TECHNICAL FIELD

The present disclosure relates generally to analyzing multimedia content, and particularly to utilization of key points in multimedia data elements.

BACKGROUND

With the abundance of multimedia data made available through various means in general and the Internet and world-wide web (WWW) in particular, there is a need for effective ways of analyzing and identifying such multimedia data. Such multimedia data may include, for example, images, graphics, video streams, video clips, video frames, photographs, images of signals, and the like.

Analyzing such multimedia content may be challenging at best due to the huge amount of information that needs to be examined. Helpful and vital data analysis becomes time intensive due to the amount of data that must be processed. As a result, data analysis may be a low priority or ignored entirely.

Even identifying multimedia content elements included in multimedia content is a challenging problem. Existing solutions may include processing, analyzing, and understanding the multimedia content based on one or more decisions. A theme of development for many such existing solutions has been to replicate the abilities of human vision by electronically perceiving and recognizing multimedia content items.

The existing solutions are limited in the ability to identify multimedia content elements that are received for the first time. In particular, many existing solutions require comparing portions of multimedia content to known multimedia content elements to identify any matching multimedia content elements. Thus, unknown multimedia content elements or multimedia content elements that have otherwise never been received before may not be successfully recognized.

Additionally, existing solutions are often highly sensitive to changes in the received multimedia content elements. Consequently, minor changes in the multimedia content due to, for example, differences during capturing, may result in otherwise identical multimedia content elements being unrecognizable. For example, taking pictures of a car at different angles (e.g., one from the rear right side and another from the front left side) may result in the car being unrecognizable by existing solutions for one or more of the pictures.

Other existing solutions rely on metadata to identify multimedia content elements. Use of such metadata typically relies on information from, e.g., users. Thus, the metadata may not be sufficiently defined to fully describe the multimedia content and, as a result, may not capture all aspects of the multimedia content. For example, a picture of a car may be associated with metadata representing a model of the car, but other pictures of the car may not be associated with metadata designating the owners.

Further, the existing solutions often include analyzing each and every pixel of the multimedia content and matching those pixels to pixels in a database. This analysis consumes a significant amount of computing resources. Further, such a complete analysis may be unnecessary, as portions of multimedia content may not contribute to the identification of elements therein. For example, images may include black edges which are not useful for identification. As another example, text that is repeated throughout an image may only be useful to identification once, and subsequent analysis of the text is not useful.

It would therefore be advantageous to provide a solution that would overcome the deficiencies of the prior art.

SUMMARY

A summary of several example embodiments of the disclosure follows. This summary is provided for the convenience of the reader to provide a basic understanding of such embodiments and does not wholly define the breadth of the disclosure. This summary is not an extensive overview of all contemplated embodiments, and is intended to neither identify key or critical elements of all embodiments nor to delineate the scope of any or all aspects. Its sole purpose is to present some concepts of one or more embodiments in a simplified form as a prelude to the more detailed description that is presented later. For convenience, the term “some embodiments” may be used herein to refer to a single embodiment or multiple embodiments of the disclosure.

Certain embodiments disclosed herein include a method for determining common patterns in multimedia data elements (MMDEs) based on key points. The method comprises: identifying a plurality of candidate key points in each of the plurality of MMDEs, wherein a size of each candidate key point is equal to a predetermined size and a scale of each candidate key point is equal to a predetermined scale; analyzing the identified candidate key points to determine a set of properties for each candidate key point; comparing the sets of properties of the plurality of candidate key points of each MMDE; selecting, for each MMDE, a plurality of key points from among the candidate key points based on the comparison; generating, based on the plurality of key points for each MMDE, a signature for the MMDE; and comparing the signatures of the plurality of MMDEs to output at least one common pattern among the plurality of MMDEs.

Certain embodiments disclosed herein also include a non-transitory computer readable medium having stored thereon instructions for causing one or more processing units to execute a method, the method comprising: identifying a plurality of candidate key points in each of the plurality of MMDEs, wherein a size of each candidate key point is equal to a predetermined size and a scale of each candidate key point is equal to a predetermined scale; analyzing the identified candidate key points to determine a set of properties for each candidate key point; comparing the sets of properties of the plurality of candidate key points of each MMDE; selecting, for each MMDE, a plurality of key points from among the candidate key points based on the comparison; generating, based on the plurality of key points for each MMDE, a signature for the MMDE; and comparing the signatures of the plurality of MMDEs to output at least one common pattern among the plurality of MMDEs.

Certain embodiments disclosed herein also include a system for determining common patterns in multimedia data elements (MMDEs) based on key points. The system comprises: a processing circuitry; and a memory, the memory containing instructions that, when executed by the processing circuitry, configure the system to: identify a plurality of candidate key points in each of the plurality of MMDEs, wherein a size of each candidate key point is equal to a predetermined size and a scale of each candidate key point is equal to a predetermined scale; analyze the identified candidate key points to determine a set of properties for each candidate key point; compare the sets of properties of the plurality of candidate key points of each MMDE; select, for each MMDE, a plurality of key points from among the candidate key points based on the comparison; generate, based on the plurality of key points for each MMDE, a signature for the MMDE; and compare the signatures of the plurality of MMDEs to output at least one common pattern among the plurality of MMDEs.

BRIEF DESCRIPTION OF THE DRAWINGS

The subject matter disclosed herein is particularly pointed out and distinctly claimed in the claims at the conclusion of the specification. The foregoing and other objects, features, and advantages of the disclosed embodiments will be apparent from the following detailed description taken in conjunction with the accompanying drawings.

FIG. 1 is a schematic diagram of a system for identifying key points in multimedia data elements according to an embodiment.

FIG. 2 is a schematic diagram of a properties generation unit according to an embodiment.

FIG. 3 is a flowchart illustrating a method for identifying key points in multimedia data elements according to an embodiment.

FIG. 4 is a flowchart illustrating a method for generating properties based on candidate key points according to an embodiment.

FIG. 5 is a flow chart illustrating a method for selecting key points from among candidate key points according to an embodiment.

FIG. 6 is an example simulation of identifying key points in an image multimedia data element.

FIG. 7 is an example data plot utilized to illustrate determining a location property of a candidate key point.

FIG. 8 is an example simulation of determining a rotation property of a candidate key point.

FIG. 9 is an example data plot utilized to illustrate determining a size property of a candidate key point.

FIG. 10 is an example data plot utilized to illustrate determining a pixilation property of a candidate key point.

FIG. 11 is an example data plot utilized to illustrate determining a key point based on an analysis of a set of properties of each of a plurality of candidate key points.

FIG. 12 is a flowchart illustrating a method for determining common patterns in multimedia data elements based on identified key points according to an embodiment.

FIG. 13 is a block diagram depicting the basic flow of information in a signature generator system.

FIG. 14 is a diagram showing the flow of patches generation, response vector generation, and signature generation in a large-scale speech-to-text system.

DETAILED DESCRIPTION

It is important to note that the embodiments disclosed herein are only examples of the many advantageous uses of the innovative teachings herein. In general, statements made in the specification of the present application do not necessarily limit any of the various claimed embodiments. Moreover, some statements may apply to some inventive features but not to others. In general, unless otherwise indicated, singular elements may be in plural and vice versa with no loss of generality. In the drawings, like numerals refer to like parts through several views.

The various disclosed embodiments include a method and system for determining common patterns in multimedia data elements based on key points in the multimedia data elements. The identified key points may be utilized to identify multimedia content elements in the multimedia data elements. A multimedia data element is analyzed to identify candidate key points. The candidate key points are analyzed to determine a set of properties for each candidate key point. Key points are selected based on the determined sets of properties. Signatures are generated for the selected key points. In some embodiments, the generated signatures may be clustered into a cluster representing the multimedia data element. The signatures or cluster of the multimedia data element may be compared to signatures of other multimedia data elements to output common patterns among the multimedia data elements.

FIG. 1 shows an example schematic diagram of a system 100 for determining common patterns based on key points in multimedia data elements (MMDEs) according to an embodiment. The system 100 includes an interface 110, a processing circuitry 120, a memory 130, a properties generator (PG) 140, a storage 150, and a signature generator 160. The memory 130 may be volatile (e.g., RAM, etc.), non-volatile (e.g., ROM, flash memory, etc.), or a combination thereof.

Key points are areas within an MMDE of predetermined size and scale that are determined by the system 100 to be the best representations of elements shown in the MMDE. A key point is an area of interest within the MMDE. Key points can be utilized to allow for efficient identification of elements shown in the MMDE by, for example, computer vision systems. Further, the key points may be utilized as representative portions of a MMDE for generating signatures thereto such that the signatures are only generated for the representative portions, thereby reducing use of computing resources and improving accuracy during signature generation.

As an example, for a picture of a cat lying on grass, portions of the picture in which the cat or part of the cat is shown may be considered of stronger interest than portions in which only grass is shown. Thus, the area in the picture showing the cat is a key point. As another example, for a picture of a sunset reflected in the ocean, portions of the picture in which both the sun and ocean appear may be considered key points, while portions featuring only the sun or the ocean may not be considered key points.

The key points may be determined based on at least one candidate key point identified in an MMDE. In an embodiment, the identified candidate key points may be selected randomly from among points in the MMDE identified during the analysis. In another embodiment, the candidate key points may be identified based on at least one predetermined key point representation rule. As a non-limiting example, a key point representation rule may include a distance threshold such that only one of any two points having a distance between the two points less than the distance threshold are is selected as a candidate key point. A key point representation rule is described herein below.

MMDEs may be received through the interface 110. The interface 110 may be, but is not limited to, a network interface. As an example, the interface 110 may be a network interface for receiving MMDEs from one or more data sources (not shown) over a network (not shown). The data sources may be, for example, servers (e.g., web servers) or other sources of data including MMDEs. Each MMDE may be, but is not limited to, an image, a graphic, a video stream, a video clip, a video frame, a photograph, and an image of signals (e.g., spectrograms, phasograms, scalograms, etc.), combinations thereof, and portions thereof.

The properties generator 140 is configured to generate a set of properties for each candidate key point. The properties are scalable measures enabling evaluation of each candidate key point as well as determination of key points from among the candidate key points. The properties may include, but are not limited to, a location of a candidate key point within an MMDE, a rotation of a candidate key point within the MMDE, a size of the candidate key point relative to the MMDE, a pixilation of the candidate key point, combinations thereof, and the like. In an embodiment, the properties generator 140 may be further configured to identify benchmarking metrics utilized for determining properties of the candidate key points. For example, for an image, benchmarking metrics may include a white color against which other colors in the image may be compared. The benchmarking metrics utilized and properties determined may be based on a type of the MMDE. For example, metrics for an image may differ from metrics for audio.

The location of the candidate key point may be represented in an XY diagram, wherein the point (0,0) represents one or more edges of the MMDE. The size of a candidate key point is a size of a multimedia content element. The rotation of the candidate key point is an angle at which a multimedia content element located at the candidate key point is tilted with respect to a baseline and may be determined respective of, for example, 8 different benchmarking metrics representing different rotations. The baseline may be further determined based on other multimedia content elements of the MMDE. The pixilation may be represented, e.g., in two rectangles (e.g., a 6×3 black rectangle and a 3×6 white rectangle).

The properties generator 140 may be further configured to store the generated properties in the storage 150. The properties generator 140 is described further herein below with respect to FIG. 2.

The signature generator 160 is configured to generate signatures for MMDEs or portions thereof. Each signature is a distinct numeric representation of an MMDE or portion thereof such that different signatures represent different features of MMDEs or portions thereof, which may be robust to noise and distortion. The signature generator 160 may be further configured to compare a plurality of signatures. The comparison may include, but is not limited to, mapping a signature vector representation into vector space, using Euclidean distance as a first approximation, and the like. This comparison may be assisted by embedding algorithms which reduce the number of dimensions in the vector space. In an embodiment, the signature generator 160 may be configured to generate signatures as described further herein below with respect to FIGS. 13 and 14.

The processing circuitry 120 is configured to receive one or more MMDEs through the interface 110 and to determine candidate key points for each of the received MMDEs. The processing circuitry 120 is further configured to cause the properties generator 140 to generate the set of properties for each determined candidate key point and to retrieve the generated sets of properties. Based on the retrieved properties, the processing circuitry 120 is configured to identify key points from among the candidate key points.

The processing circuitry 120 is typically coupled to the memory 130. The processing circuitry 120 may comprise or be a component of a processor (not shown) or an array of processors coupled to the memory 130. The memory 130 contains instructions that can be executed by the processing circuitry 120. The instructions, when executed by the processing circuitry 120, cause the processing circuitry 120 to perform the various functions described herein. The one or more processors may be implemented with any combination of general-purpose microprocessors, multi-core processors, microcontrollers, digital signal processors (DSPs), field programmable gate array (FPGAs), programmable logic devices (PLDs), controllers, state machines, gated logic, discrete hardware components, dedicated hardware finite state machines, or any other suitable entities that can perform calculations or other manipulations of information.

The processing circuitry 120 may also include machine-readable media for storing software. Software shall be construed broadly to mean any type of instructions, whether referred to as software, firmware, middleware, microcode, hardware description language, or otherwise. Instructions may include code (e.g., in source code format, binary code format, executable code format, or any other suitable format of code). The instructions, when executed by the one or more processors, cause the processing system to perform the various functions described herein.

In another embodiment, the processing circuitry 120 can be realized as an array of computational cores, each core having properties that are at least partly statistically independent from other cores of the plurality of computational cores. The array of computational cores may be initialized by the signature generator 160 to generate signatures. Such cores are generated or otherwise configured to obtain maximal independence, i.e., the projection from a signal space should generate a maximal pair-wise distance between any two cores' projections into a high-dimensional space. Further, the cores are optimally designed for the type of signals, i.e., the cores should be maximally sensitive to the spatio-temporal structure of the injected signal, for example, and in particular, sensitive to local correlations in time and space. Thus, in some cases a core represents a dynamic system, such as in state space, phase space, edge of chaos, etc., which is uniquely used herein to exploit their maximal computational power. In addition, the computational cores are optimally designed with regard to invariance to a set of signal distortions, of interest in relevant applications. A detailed description of processes for generating, configuring, and operating any array of computational cores is discussed in more detail in U.S. Pat. No. 8,655,801 assigned to the common assignee, which is hereby incorporated by reference for all the useful information it contains.

It should be understood that the embodiments disclosed herein are not limited to the specific architecture illustrated in FIG. 1, and that other architectures may be equally used without departing from the scope of the disclosed embodiments. Moreover, in an embodiment, there may be a plurality of systems 100 operating as described hereinabove and configured to either have one as a standby, to share the load between them, or to split the functions between them.

FIG. 2 is an example flow diagram illustrating image processing by the properties generator 140 according to an embodiment. In the example embodiment shown in FIG. 2, the properties generator 140 includes a location determination circuit (LDC) 141, a rotation determination circuit (RDC) 142, a size determination circuit (SDC) 143, and a pixilation determination circuit (PDC) 144. In an optional implementation, the properties generator 140 also includes a comparator 145. Further, in the example embodiment shown in FIG. 2, a MMDE is processed in the following order: by the location determination circuit 141, by the rotation determination circuit 142, by the size determination circuit 143, by the pixilation determination circuit 144, and by the comparator 145.

The location determination circuit 141 is configured to determine a location of a candidate key point in a MMDE. The location may be relative to the MMDE. To this end, when the MMDE is an image, the location may be expressed as a pair of, e.g., X and Y coordinates (X,Y). The origin (0,0) may be any point in the MMDE. As a non-limiting example, the origin may be at the bottom left corner of the MMDE such that points in the MMDE are at coordinates (0,0), (100,150), and any coordinates in between such as, but not limited to, (0,90), (50,0), (75,75), (99,149), (80,120), and so on. When the MMDE is audio, the location may be expressed as a moment or period of time in the audio file. For example, for an audio clip that is 5 minutes (300 seconds) long, the location may be, but is not limited to, 0 seconds (start), 30 seconds, 100 seconds, 267 seconds, 150.2 seconds, 300 seconds (end), and the like.

The rotation determination circuit 142 is configured to determine a rotation of candidate key points in MMDEs. As a non-limiting example, the rotation determination circuit 142 is configured to identify edges of MMDEs, thereby enabling matching of the rotation of the MMDEs based on the respective edges thereof. The size determination circuit 143 is configured to determine a size of candidate key points in MMDEs. The pixilation determination circuit is configured to 144 determines a pixilation of candidate key points in MMDEs. As a non-limiting example, in an image of a couple hugging in front of the Eiffel tower, the pixilation of the portion of the image showing the couple is higher than the pixilation of the portion of the image showing the Eiffel tower in the background. Each of the location determination circuit 141, the rotation determination circuit 142, the size determination circuit is configured to 143, and the pixilation determination circuit 144 may determine its respective properties based on characteristics of elements in the MMDE. To this end, each circuit, 142, 143, or 144 may be configured to identify at least one benchmarking metric based on the MMDE and to compare elements in the MMDE to the benchmarking metric.

Each benchmarking metric may be a metric representing a particular rotation, size, or pixilation of an MMDE, and may be utilized as a point of comparison by, for example, the rotation determination circuit 142, the size determination circuit 143, or the pixilation determination circuit 144, respectively. To this end, each of the rotation, size, and pixilation of a MMDE may be determined relative to at least one corresponding benchmarking metric. As a non-limiting example, if a text element in an image is identified as being in a particular character set (e.g., letters of the English alphabet), the rotation determination circuit 142 may be configured to determine a rotation of the text element with respect to a benchmarking metric text element in the same character set (e.g., a horizontally oriented text element using English alphabet letters).

In an embodiment, the properties generator 140 may include a comparison unit 145. The comparison unit 145 compares a set of properties of each candidate key point. The comparison may be utilized to determine whether a candidate key point should be selected as a key point. To this end, the comparator 145 may be configured to compare scores of properties of the same type (e.g., scores for locations of different candidate key points, scores for rotations of different candidate key points, and the like).

In an embodiment, each, some, or all of the location determination circuit 141, the rotation determination circuit 142, the size determination circuit 143, the pixilation determination circuit (PDC) 144, and the comparator 145 may comprise or be a component of a processor (not shown) or an array of processors. Examples for such processor or processors are provided above.

It should be noted that the flow diagram shown in FIG. 2 is merely an example and does not limit any of the disclosed embodiments. In particular, a MMDE may be processed by any of the location determination circuit 141, the rotation determination circuit 142, the size determination circuit 143, and the pixilation determination circuit 144, either in series or in parallel, and may be processed by each circuit in any order. As a non-limiting example, the MMDE may be processed by the location determination circuit 141, the rotation determination circuit 142, the size determination circuit 143, and the pixilation determination circuit 144 simultaneously. As another non-limiting example, the MMDE may be processed by the circuits in the following order: by the rotation determination circuit 142, by the pixilation determination circuit 144, by the size determination circuit 143, and by the location determination circuit 141. Additionally, other circuits for determining properties of MMDEs (not shown), such as a color determination circuit, may be equally used in addition to or instead of any of the circuits 141, 142, 143, or 144.

FIG. 3 is an example flowchart 300 illustrating a method for identifying key points in a MMDE according to an embodiment. In an embodiment, the method may be performed by the system 100.

At S310, a MMDE is received. The MMDE may be received via an interface (e.g., the interface 110).

At S320 the MMDE is analyzed to identify candidate key points. In an embodiment, S320 may include image-based recognition of the MMDE. In a further embodiment, the image-based recognition may begin at the edges of the MMDE and continue to the center. As an example, if the MMDE is an image, the analysis may begin at the outermost points in the image. As another example, if the MMDE is audio, the analysis may begin at the beginning and end times for the audio.

In an embodiment, the identified candidate key points may be selected randomly from among points in the MMDE identified during the analysis. In another embodiment, the candidate key points may be identified based on at least one predetermined key point representation rule. As an example, a key point representation rule may include a distance threshold (e.g., a distance between points in an image or video, a length of time in audio, etc.). If two points in an MMDE are separated by a distance less than the distance threshold, only one of the points may be identified as a candidate key point.

At S330, a set of properties is determined for each identified candidate key point. Determination of properties for candidate key points is described further herein below with respect to FIG. 4.

At S340, the properties for each candidate key point are compared. In an embodiment, comparing the properties further includes determining a property score for each property of each candidate key point. The property scores may be determined based on comparison of characteristics of each property such as, but not limited to, intensity, distance from a center point of the MMDE, color, angle of rotation, a combination thereof, and the like. The property scores may be determined further based on benchmarking metrics for such characteristics. In an embodiment, higher property scores indicating a greater likely significance of the candidate key point. As an example, the location scores for a particular candidate point may be 3, 7, and 8, respectively, with 1 representing the lowest likelihood of significance (e.g., toward the outer edges of the MMDE) and 10 representing the highest likelihood of significance (e.g., closest to the center of the MMDE).

At S350, key points are selected from among the identified candidate key points. The key points may be selected based on the determined sets of properties via, e.g., comparison of the properties' respective scores. Selecting key points among candidate key points is described further herein below with respect to FIG. 5.

At optional S360, it is checked whether additional key points are required and, if so, execution continues with S320; otherwise, execution terminates. In an embodiment, upon selecting a key point in a particular area of the MMDE, additional key points may be checked for within, or in proximity to, the area of the key point.

FIG. 4 is an example flowchart S340 illustrating a method for determining a set of properties for candidate key points in an MMDE according to an embodiment. In an embodiment, each of the properties may be determined by comparing benchmarking metrics to one or more characteristics of the candidate key point.

At S410, a location of a candidate key point is determined. The location of the candidate key point may be determined by identifying a center point of the MMDE and determining a distance from the center point to the candidate key point. At S420, a rotation of the candidate key point may be determined. The rotation may be determined based on edges identified in the MMDE. At S430, a size of the candidate key point may be determined. At S440, a pixilation of the candidate key point may be determined.

At S450, it may be determined whether properties of additional candidate key points are required and, if so, execution continues with S410; otherwise execution terminates. In an embodiment, the determination may be based on an MMDE identification rule. The MMDE identification rule indicates at least one condition for successful identification of multimedia content elements and may be based on, but not limited to, an event (e.g., identification of a concept related to the MMDE), a threshold (e.g., a number of sets of properties for candidate key points), a combination thereof, and the like. To this end, in an embodiment in which identification of a concept related to the MMDE is indicated by the MMDE identification rule, S450 may further include determining whether a concept can be identified based on the properties determined thus far.

It should be noted that FIG. 4 is described herein above with respect to location, rotation, size, and pixilation properties merely for simplicity purposes and without limitation on the disclosed embodiments. More, fewer, or other properties may be utilized without departing from the scope of the disclosure. As an example, a color property for an image may also be determined. As another example, a volume property for an audio file may be determined, and the rotation and pixilation properties may not be determined for the audio file.

FIG. 5 is an example flowchart S350 illustrating a method for selecting key points in a MMDE according to an embodiment.

At S510, sets of properties of candidate key points in the MMDE are obtained. The sets of properties for each candidate key point may include the properties determined as described herein above with respect to FIG. 4.

At S520, the sets of properties are compared to identify relatively high sets of properties. The relatively high sets of properties are identified to determine the most descriptive candidate key points. In an embodiment, S520 includes determining a property score for each property. Each property score may be determined based on relative values for properties of the candidate key points. In a further embodiment, S520 may also include determining an average property score for properties of each set of properties. In yet a further embodiment, relatively high sets of properties may be sets of properties having average property scores above a predetermined threshold.

At optional S530, at least one budget parameter may be retrieved. The budget parameter is a quantitative limitation on the maximum amount of key points that may be selected for the MMDE and is typically utilized to ensure efficient key point identification by restricting the number of key points that need to be identified, thereby conserving computing resources. The budget may be the same for all MMDEs, may differ for different types of MMDEs, and the like. In an embodiment, the budget may be retrieved from the storage unit 150.

At S540, key points to be selected are determined based on the comparison. The number of key points determined may be limited based on the budget.

FIG. 6 is an example simulation of a selection of candidate key points in a MMDE. In the example simulation of FIG. 6, an image 600 includes a cat. The image 600 may be received via an interface and analyzed. Based on the analysis, candidate key points 610-1 through 610-4 are identified. A set of properties is generated for each of the candidate key points 610. The sets of properties are compared. For example, the location property of candidate key point 610-1 is relatively low as compared to candidate key point 610-3 because it is closer to the center of the image 600. Additionally, the pixilation property of candidate key point 610-2 may be relatively low as compared to candidate key point 610-4.

FIG. 7 is an example data plot 700 utilized to illustrate determining a location property of a candidate key point. The example data plot 700 includes a set of data points 710 and a subset of data points 720 among the set of data points 710. Location properties are determined for candidate key points. The location properties may be displayed in the data plot 700 as an XY graph where the data point (0,0) represents a point on an edge of a MMDE and where the points farthest from the data point (0,0) represent the points closest to the center of the MMDE. In the example data plot 700, each point (x,y) represents, e.g., a horizontal and vertical distance, respectively, of the candidate key point to the center of the MMDE when the MMDE is an image. In other example data plots, each point (x,y) may represent, e.g., a horizontal and vertical distance, respectively, of the candidate key point to one or more of the edges of the image MMDE. For other types of multimedia content elements (e.g., video, audio, etc.), each of the X-axis and the Y-axis may represent metrics such as, but not limited to, amount of time from half of the total time of the audio or video, a horizontal or vertical distance to or from a center of an image-based portion of the video, a distance of a line to or from the center of the image-based portion of the video, and the like.

The candidate key points with the strongest responses (i.e., location properties) may be selected. The strongest response key points may be determined by comparing the location properties among the candidate key points 710 and assigning a location score to each of the candidate key points 710. As an example, the strongest response candidate key points are associated with points of the subset 720.

FIG. 8 is an example simulation 800 of determining a rotation property of a candidate key point. The simulation 800 is based on an image 810 of a superhero character standing in front of a city background and includes candidate key points 820 and 830. A rotation property may be determined for each of the candidate key points 820 and 830. The determination may include comparing rotations among the candidate key points and determining a rotation scale based on the comparison. The determination may further include analyzing the rotation of each candidate key point respective of the rotation scale and assigning a rotation score to each candidate key point. As an example, the candidate key point 820 may be assigned a lower rotation score than that of candidate key point 830 because the rotation of the candidate key point 820 (i.e., a point on the side of a vertically oriented building) is less than the rotation of the candidate key point 830 (i.e., a point on an angle of a bent elbow).

FIG. 9 is an example data plot 900 utilized to illustrate determining size properties of candidate key points. The example data plot 900 includes key points 910 and a data point cluster 920. Each of the data points 910 represents a size property of a candidate key point, with the data point (0,0) representing the lowest size property and the data points of the key point cluster 920 representing the highest size properties among candidate key points. Thus, the largest candidate key points are associated with a higher size score and are the candidate key points associated with the data point cluster 920.

FIG. 10 is an example data plot 1000 utilized to illustrate determining pixilation properties of candidate key points. The pixilation properties represent visual quality of the MMCE as determined based on, e.g., resolution. The example data plot 1000 includes key points 1010 and a data point cluster 1020 where the data point (0,0) represents the lowest pixilation among candidate key points and the candidate key points having the highest pixilation are represented by data points of the cluster 1020. Key points with the higher pixilation are assigned with a higher pixilation score.

FIG. 11 is an example data plot 1100 utilized to illustrate selecting key points based on an analysis of a set of properties of each of a plurality of candidate key points. The example data plot 1100 includes points 1110 and 1120. Each point 1110 and 1120 represents a candidate key point having a set of properties. In an example implementation, the set of properties may include 2 or 4 properties. The set of properties for each candidate key point is analyzed to determine which candidate key points are to be utilized as key points. As an example, the candidate key point may be assigned da score for each property, and one or more candidate key points having a highest average score from among a plurality of candidate key points may be selected as key points.

FIG. 12 is an example flowchart 1200 illustrating a method for determining common patterns in MMDEs based on key points identified therein according to an embodiment. In an embodiment, the method may be performed by the system 100, FIG. 1.

At S1210, a plurality of MMDEs is obtained. The plurality of MMDEs may be received (e.g., from a user device), retrieved (e.g., from a storage), or a combination thereof.

At S1220, a plurality of key points in each obtained MMDE is identified. In an embodiment, the plurality of key points for each MMDE is identified as described further herein above with respect to FIGS. 3 through 5.

At S1230, based on the identified key points, a signature is generated for each of the plurality of MMDEs. The signature for each MMDE is a distinct numeric representation of the MMDE, and may be robust to noise and distortion. In an embodiment, generating the signature for each MMDE may include generating a signature for each key point in the MMDE and clustering the generated key point signatures to create the signature of the MMDE. Such key point-based signature generation results in only generating signatures for select portions of the MMDE that best represent the MMDE and, therefore, may reduce use of computing resources for signature generation, result in more accurate signatures (i.e., signatures that better represent unique aspects of the MMDE), or both.

In an embodiment, each signature may be generated by a signature generator system. The signature generator system includes a plurality of at least partially statistically independent computational cores, wherein the properties of each computational core are set independently of the properties of the other cores. A detailed description of such a signature generation process is provided in the above-mentioned U.S. Pat. No. 8,655,801.

At S1240, based on the generated signatures, at least one common pattern among the plurality of multimedia content elements is output. In an embodiment, S1240 includes matching between signatures of the plurality of multimedia content elements to identify portions of at least two signatures matching above a predetermined threshold. The matching identified portions of signatures may be utilized to generate the at least one common pattern. The at least one common pattern may be or may include a signature representing common features of two or more MMDEs. To this end, in a further embodiment, S1240 may include selecting a representative signature from among two or more matching portions of signatures to be utilized as a common pattern.

The determined common patterns may be hidden patterns, i.e., patterns that are not revealed using other methods (e.g., based on comparison of metadata associated with the MMDEs). The determined common patterns may be utilized to, e.g., organize the plurality of MMDEs, generate similarity statistics, identify anomalies, and the like. For example, the MMDEs may be grouped with respect to the determined common patterns.

FIGS. 13 and 14 illustrate an example for generation of signatures for the multimedia content elements according to an embodiment. An example high-level description of the process for large scale matching is depicted in FIG. 13. In this example, the matching is for a video content.

Video content segments 2 from a Master database (DB) 6 and a Target DB 1 are processed in parallel by a large number of independent computational Cores 3 that constitute an architecture for generating the Signatures (hereinafter the “Architecture”). Further details on the computational Cores generation are provided below. The independent Cores 3 generate a database of Robust Signatures and Signatures 4 for Target content-segments 5 and a database of Robust Signatures and Signatures 7 for Master content-segments 8. An exemplary and non-limiting process of signature generation for an audio component is shown in detail in FIG. 14. Finally, Target Robust Signatures and/or Signatures are effectively matched, by a matching algorithm 9, to Master Robust Signatures and/or Signatures database to find all matches between the two databases.

To demonstrate an example of the signature generation process, it is assumed, merely for the sake of simplicity and without limitation on the generality of the disclosed embodiments, that the signatures are based on a single frame, leading to certain simplification of the computational cores generation. The Matching System is extensible for signatures generation capturing the dynamics in-between the frames.

The Signatures' generation process is now described with reference to FIG. 14. The first step in the process of signatures generation from a given speech-segment is to breakdown the speech-segment to K patches 14 of random length P and random position within the speech segment 12. The breakdown is performed by the patch generator component 21. The value of the number of patches K, random length P and random position parameters is determined based on optimization, considering the tradeoff between accuracy rate and the number of fast matches required in the flow process of the context server 130 and SGS 140. Thereafter, all the K patches are injected in parallel into all computational Cores 3 to generate K response vectors 22, which are fed into a signature generator system 23 to produce a database of Robust Signatures and Signatures 4.

In order to generate Robust Signatures, i.e., Signatures that are robust to additive noise L (where L is an integer equal to or greater than 1) by the Computational Cores 3 a frame ‘i’ is injected into all the Cores 3. Then, Cores 3 generate two binary response vectors: {right arrow over (S)} which is a Signature vector, and {right arrow over (RS)} which is a Robust Signature vector.

For generation of signatures robust to additive noise, such as White-Gaussian-Noise, scratch, etc., but not robust to distortions, such as crop, shift and rotation, etc., a core Ci={ni} (1≤i≤L) may consist of a single leaky integrate-to-threshold unit (LTU) node or more nodes. The node ni equations are:

$V_i=\sum _j{w}_{\mathrm{ij}}{k}_j$ $n_i=\theta \left(\mathrm{Vi}-{\mathrm{Th}}_x\right)$

where, θ is a Heaviside step function; w_ij is a coupling node unit (CNU) between node i and image component j (for example, grayscale value of a certain pixel j); kj is an image component ‘j’ (for example, grayscale value of a certain pixel j); Thx is a constant Threshold value, where ‘x’ is ‘S’ for Signature and ‘RS’ for Robust Signature; and Vi is a Coupling Node Value.

The Threshold values Thx are set differently for Signature generation and for Robust Signature generation. For example, for a certain distribution of Vi values (for the set of nodes), the thresholds for Signature (Th_s) and Robust Signature (Th_Rs) are set apart, after optimization, according to at least one or more of the following criteria:

1: ForV_i>Th_RS 1−p(V>Th_S)−1−(1−ε)¹<<1

• • i.e., given that/nodes (cores) constitute a Robust Signature of a certain image I, the probability that not all of these I nodes will belong to the Signature of same, but noisy image, Ĩ is sufficiently low (according to a system's specified accuracy).

2:p(V_i>Th_RS)≈I/L

• • i.e., approximately I out of the total L nodes can be found to generate a Robust Signature according to the above definition.
  • 3: Both Robust Signature and Signature are generated for certain frame i.

It should be understood that the generation of a signature is unidirectional, and typically yields lossless compression, where the characteristics of the compressed data are maintained but the uncompressed data cannot be reconstructed. Therefore, a signature can be used for the purpose of comparison to another signature without the need of comparison to the original data. The detailed description of the Signature generation can be found in U.S. Pat. Nos. 8,326,775 and 8,312,031, assigned to the common assignee, which are hereby incorporated by reference for all the useful information they contain.

A Computational Core generation is a process of definition, selection, and tuning of the parameters of the cores for a certain realization in a specific system and application. The process is based on several design considerations, such as:

• • (a) The Cores should be designed so as to obtain maximal independence, i.e., the projection from a signal space should generate a maximal pair-wise distance between any two cores' projections into a high-dimensional space.
  • (b) The Cores should be optimally designed for the type of signals, i.e., the Cores should be maximally sensitive to the spatio-temporal structure of the injected signal, for example, and in particular, sensitive to local correlations in time and space. Thus, in some cases a core represents a dynamic system, such as in state space, phase space, edge of chaos, etc., which is uniquely used herein to exploit their maximal computational power.
  • (c) The Cores should be optimally designed with regard to invariance to a set of signal distortions, of interest in relevant applications.

A detailed description of the Computational Core generation and the process for configuring such cores is discussed in more detail in the above-referenced U.S. Pat. No. 8,655,801.

The various embodiments disclosed herein can be implemented as hardware, firmware, software, or any combination thereof. Moreover, the software is preferably implemented as an application program tangibly embodied on a program storage unit or computer readable medium consisting of parts, or of certain devices and/or a combination of devices. The application program may be uploaded to, and executed by, a machine comprising any suitable architecture. Preferably, the machine is implemented on a computer platform having hardware such as one or more central processing units (“CPUs”), a memory, and input/output interfaces. The computer platform may also include an operating system and microinstruction code. The various processes and functions described herein may be either part of the microinstruction code or part of the application program, or any combination thereof, which may be executed by a CPU, whether or not such a computer or processor is explicitly shown. In addition, various other peripheral units may be connected to the computer platform such as an additional data storage unit and a printing unit. Furthermore, a non-transitory computer readable medium is any computer readable medium except for a transitory propagating signal.

All examples and conditional language recited herein are intended for pedagogical purposes to aid the reader in understanding the principles of the disclosed embodiment and the concepts contributed by the inventor to furthering the art, and are to be construed as being without limitation to such specifically recited examples and conditions. Moreover, all statements herein reciting principles, aspects, and embodiments of the disclosed embodiments, as well as specific examples thereof, are intended to encompass both structural and functional equivalents thereof. Additionally, it is intended that such equivalents include both currently known equivalents as well as equivalents developed in the future, i.e., any elements developed that perform the same function, regardless of structure.

Claims

1. A method for determining common patterns in a plurality of multimedia data elements (MMDEs), comprising: identifying, by an array of computational cores of a processor, a plurality of candidate key points in each of the plurality of MMDEs, wherein a size of each candidate key point is equal to a predetermined size and a scale of each candidate key point is equal to a predetermined scale;analyzing the plurality of candidate key points in each MMDE of the plurality of MMDEs to determine a set of properties for each candidate key point;for each MMDE of the plurality of MMDEs, comparing the sets of properties of the plurality of candidate key points of the respective MMDE;selecting, for each MMDE, a plurality of key points from among the candidate key points in the respective MMDE, based on the comparing;generating, based on the plurality of key points for each MMDE, a signature for each MMDE to provide generated signatures; andcomparing the generated signatures to output at least one common pattern among the plurality of MMDEs; andwherein the generating of the signature for each MMDE further comprises: generating a key point signature for each selected key point of the respective MMDE; andclustering the generated key point signatures to generate the signature for the respective MMDE.

2. The method of claim 1, wherein each common pattern of the at least one common pattern is a signature representing features that are common to at least two of the plurality of MMDEs.

3. The method of claim 2, wherein the comparing of the generated signatures to output the at least one common pattern further comprises: matching the generated signatures to determine at least one set of matching portions of the generated signatures ; and generating, based on the determined at least one set of matching portions, the at least one common pattern.

4. The method of claim 1, further comprising: identifying, in each MMDE, a plurality of points having the predetermined size and the predetermined scale, wherein the candidate key points are selected from among the identified plurality of points.

5. The method of claim 4, wherein the candidate key points are identified based on at least one key point representation rule.

6. The method of claim 1, wherein each set of properties comprises pixilation and at least one of: location, rotation, size, and color .

7. The method according to claim 6 wherein each set of properties comprises the pixilation, location, rotation and size.

8. The method according to claim 1 comprising: mapping each signature generated for the plurality of MMDEs into a vector space; and reducing a number of dimensions in the vector space; wherein the comparing signatures generated for the plurality of MMDEs follows the reducing of the dimensions.

9. The method according to claim 1 wherein each computational core comprises properties that are at least partly statistically independent from properties of other computational cores of the plurality of computational cores.

10. A method for determining common patterns in a plurality of multimedia data elements (MMDEs), comprising: identifying, by an array of computational cores of a processor, a plurality of candidate key points in each of the plurality of MMDEs, wherein a size of each candidate key point is equal to a predetermined size and a scale of each candidate key point is equal to a predetermined scale;analyzing the plurality of candidate key points in each MMDE of the plurality of MMDEs to determine a set of properties for each candidate key point;for each MMDE of the plurality of MMDEs, comparing the sets of properties of the plurality of candidate key points of the respective MMDE;selecting, for each MMDE, a plurality of key points from among the candidate key points in the respective MMDE, based on the comparison;generating, based on the plurality of key points for each MMDE, a signature for each respective MMDE;comparing signatures generated for the plurality of MMDEs to output at least one common pattern among the plurality of MMDEs; andidentifying, in each MMDE, a plurality of points of the respective MMDE having the predetermined size and the predetermined scale, wherein the candidate key points are selected randomly from among the identified plurality of points of the respective MMDE.

11. The method according to claim 10 wherein the generating of the signature for each MMDE further comprises: generating a signature to each selected key point of the MMDE; and clustering the generated key point signatures to create the signature for the MMDE.

12. A method for determining common patterns in a plurality of multimedia data elements (MMDEs), comprising: identifying, by an array of computational cores of a processor, a plurality of candidate key points in each of the plurality of MMDEs, wherein a size of each candidate key point is equal to a predetermined size and a scale of each candidate key point is equal to a predetermined scale;analyzing the plurality of candidate key points in each MMDE of the plurality of MMDEs to determine a set of properties for each candidate key point;for each MMDE of the plurality of MMDEs, comparing the sets of properties of the plurality of candidate key points of the respective MMDE;selecting, for each MMDE, a plurality of key points from among the candidate key points in the respective MMDE, based on the comparison;generating, based on the plurality of key points for each MMDE, a signature for each respective MMDE; andcomparing signatures generated for the plurality of MMDEs to output at least one common pattern among the plurality of MMDEs;wherein the comparing of the sets of properties of the plurality of candidate key points further comprises:determining a score for each property of each set of properties for each candidate key point, wherein each score is determined by comparing at least one characteristic of the candidate key point to at least one characteristic of each other candidate key point; anddetermining, based on the determined scores, the plurality of key points.

13. The method according to claim 12 wherein the determining of the score for each property of each set of properties for each candidate key point is responsive to a distance between the candidate key score and a center of an MMDE that comprises the candidate key point.

14. A non-transitory computer readable medium having stored thereon instructions for causing one or more processing units to execute a method, the method comprising: identifying, via an array of computational cores of a processor of a computer vision system, a plurality of candidate key points in each of the plurality of MMDEs, wherein a size of each candidate key point is equal to a predetermined size, wherein a scale of each candidate key point is equal to a predetermined scale;analyzing the plurality of candidate key points in each MMDE of the plurality of MMDEs to determine a set of properties for each candidate key point of the respective MMDE;for each MMDE of the plurality of MMDEs, comparing the sets of properties of the plurality of candidate key points of the respective MMDE;selecting, for each MMDE, a plurality of key points from among the candidate key points in the respective MMDE, based on the comparing;generating, based on the plurality of key points for each MMDE, a signature for each MMDE;wherein the generating of a signature for each MMDE comprises:generating a signature to each selected key point of the respective MMDE; andclustering the generated key point signatures; andwherein the method further comprises: comparing the signatures of the plurality of MMDEs to output at least one common pattern among the plurality of MMDEs.

15. A system for identifying key points in a multimedia data element (MMDE), comprising: a processor; anda memory, the memory containing instructions that, when executed by the processing circuitry, configure the system to:identify, by an array of computational cores of the processing circuitry, a plurality of candidate key points in each of the plurality of MMDEs, wherein a size of each candidate key point is equal to a predetermined size, wherein a scale of each candidate key point is equal to a predetermined scale;analyze the identified candidate key points to determine a set of properties for each candidate key point;compare the sets of properties of the plurality of candidate key points of each MMDE; select, for each MMDE, a plurality of key points from among the candidate key points based on the comparison;generate, based on the plurality of key points for each MMDE, a signature for the respective MMDE; andcompare the signatures of the plurality of MMDEs to output at least one common pattern among the plurality of MMDEs;wherein the system is configured to generate a signature for an MMDE by generating a signature to each selected key point of the respective MMDE; and clustering the generated key point signatures of the respective MMCE .

16. The system of claim 15, wherein each common pattern of the at least one common pattern is a signature representing features that are common to at least two of the plurality of MMDEs.

17. The system of claim 16, wherein the system is further configured to: match the signatures generated for the plurality of MMDEs to determine at least one set of matching portions of the signatures generated for the plurality of MMDEs; and generate, based on the determined at least one set of matching portions, the at least one common pattern.

18. The system of claim 15, wherein the system is further configured to: identify, in each MMDE, a plurality of points having the predetermined size and the predetermined scale, wherein the candidate key points are selected from among the identified plurality of points.

19. The system of claim 18, wherein the candidate key points are identified based on at least one key point representation rule.

20. The system of claim 15, wherein each set of properties comprises pixilation and at least one of: location, rotation, size, and color.