Patent Application
20150169960
The present disclosure relates to generating index data used in devices such as video players.
Modern users have many options to view audio/video programming. Home media systems can include a television, home theater audio system, a set top box and digital audio and/or A/V player. The user typically is provided one or more remote control devices that respond to direct user interactions such as buttons, keys or a touch screen to control the functions and features of the device.
Audio/video content is also available via a personal computer, smartphone or other device. Such devices are typically controlled via a buttons, keys, a mouse or other pointing device or a touch screen.
To address these and other issues and to further enhance the user experience, video processing system 102 includes a pattern recognition module 125 that creates index data 115 that can be used by a video player 114 that operates in response to user commands received via user interface 118 to receive the processed video signal 112 and to decode or otherwise process the processed video signal for display on the display device 116. In particular, the pattern recognition module 125 generates index data 115 describing content of an image sequence that is time-coded to the image sequence. For example, the pattern recognition module 125 can operate via clustering, syntactic pattern recognition, template analysis or other image, video or audio recognition techniques to recognize the content contained in the plurality of shots/scenes or other segments and to generate index data 115 that identifies or otherwise indicates the content.
In an embodiment, the pattern recognition module 125 generates the index data 115 based on color histogram data and further based on audio data and other image data such as object shapes, textures, and other patterns. For digital video, a color histogram is a representation of the distribution of colors in the frame(s). It represents the number of pixels that have same color or color range. The color histogram can be built for any kind of color space such as Monochrome, RGB, YUV or HSV. Each space has its feature and certain application scope. Like other kinds of histograms, the color histogram is a statistic that can be viewed as an approximation of an underlying continuous distribution of colors values. Thus the color histogram is relatively invariant with camera transformation. The size of color histogram is decided only by the color space configuration, which makes it provide a compact summarization of the video in spite of pixel number. For all the above reasons, color histogram is a good low-level feature for video content analysis.
The pattern recognition module 125 can generate index data 115 that indicates the content present and/or its characteristics, associated with video segments. Index data 115 can be delineated by individual images in the image sequence, shots, scenes, a group of pictures (GOP) or other time periods corresponding to a particular event or action. For example, index data 115 can delineate the start and stop of a play that includes a touchdown in a football game or a hit in baseball game.
In an embodiment, the index data 115 includes a database of metadata items that can grow or shrink dynamically. The database can store unique identifiers that correspond to particular metadata that identify content of the video in a time synchronized fashion. In this fashion, the index data 115 can include metadata that indicate content by the presence of objects, places, persons, and other things in delineated segments of the processed video signal 112. These metadata identifiers can be stored at either a certain event (start of a scene change, shot transition or start of a new Group of Pictures encoding) or a certain time interval (e.g.: every 1 second) or it can be done at every picture. An example of such metadata could be “Sunrise” meaning that particular video content is related to or shows a sunrise. Index data 115 can include song titles delineated by the start and stop of music, the appearance and exit of a certain person, place or object in one or more video segments.
The index data 115 can be used by the video player 114 to search, annotate, and/or navigate video content in a processed video signal 112 in a non-linear, non-contiguous, multilayer and/or other non-traditional fashion. If a user wishes to see sunrises, he/she can interact with a user interface 118 to and search index data 115 and quickly watch sunrise scenes in a particular movie or each of his movies on a segment-by-segment basis. In a similar fashion, if he wishes to see “Gandalf riding a horse”, index data 115 can be searched to locate a first segment when Gandalf is riding a horse, and a next-segment in this search results would be the next instance of Gandalf with a horse, etc. The user can navigate one or more video programs in this fashion, reviewing scenes with Gandalf riding a horse, until he/she finds a desired scene.
While the processed video signal 112 and index data 115 are shown separately, in an embodiment, the index data 115 can be included in the processed video signal 112, for example, with other metadata of the processed video signal. Further, while the video processing system 102 and the video player 114 are shown as separate devices, in other embodiments, the video processing system 102 and the video player can be implemented in the same device, such as a personal computer, tablet, smartphone, or other device. Further examples of the video processing system 102 and video player 114 including several optional functions and features are presented in conjunctions with
Video encoding/decoding and pattern recognition are both computational complex tasks, especially when performed on high resolution videos. Some temporal and spatial information, such as motion vectors and statistical information of blocks and shot segmentation are useful for both tasks. So if the two tasks are developed together, they can share information and economize on the efforts needed to implement these tasks. As previously described, the pattern recognition module 125 generates index data 115 describing content of image sequence of the processed video signal 112 that is time-coded to the image sequence. In particular, the pattern recognition module 125 generates the index data 115 based on coding feedback data from the video codec that includes color histogram data.
In an embodiment, the video codec 103 generates the color histogram data in conjunction with the processing of the image sequence and optionally other forms of coding feedback data. For example, the video codec 103 can generate shot transition data that identifies the temporal segments in the video signal corresponding to a plurality of shots. The pattern recognition module 125 can generates the index data 115 based on shot transition data to identify temporal segments in the video signal corresponding to the plurality of shots.
In an embodiment of the present disclosure, the video signals 110 can include a broadcast video signal, such as a television signal, high definition televisions signal, enhanced high definition television signal or other broadcast video signal that has been transmitted over a wireless medium, either directly or through one or more satellites or other relay stations or through a cable network, optical network or other transmission network. In addition, the video signals 110 can be generated from a stored video file, played back from a recording medium such as a magnetic tape, magnetic disk or optical disk, and can include a streaming video signal that is transmitted over a public or private network such as a local area network, wide area network, metropolitan area network or the Internet.
Video signal 110 and processed video signal 112 can each be differing ones of an analog audio/video (A/V) signal that is formatted in any of a number of analog video formats including National Television Systems Committee (NTSC), Phase Alternating Line (PAL) or Sequentiel Couleur Avec Memoire (SECAM). The video signal 110 and/or processed video signal 112 can each be a digital audio/video signal in an uncompressed digital audio/video format such as high-definition multimedia interface (HDMI) formatted data, International Telecommunications Union recommendation BT.656 formatted data, inter-integrated circuit sound (I2S) formatted data, and/or other digital A/V data formats.
The video signal 110 and/or processed video signal 112 can each be a digital video signal in a compressed digital video format such as H.264, MPEG-4 Part 10 Advanced Video Coding (AVC) or other digital format such as a Moving Picture Experts Group (MPEG) format (such as MPEG1, MPEG2 or MPEG4), Quicktime format, Real Media format, Windows Media Video (WMV) or Audio Video Interleave (AVI), or another digital video format, either standard or proprietary. When video signal 110 is received as digital video and/or processed video signal 112 is produced in a digital video format, the digital video signal may be optionally encrypted, may include corresponding audio and may be formatted for transport via one or more container formats.
Examples of such container formats are encrypted Internet Protocol (IP) packets such as used in IP TV, Digital Transmission Content Protection (DTCP), etc. In this case the payload of IP packets contain several transport stream (TS) packets and the entire payload of the IP packet is encrypted. Other examples of container formats include encrypted TS streams used in Satellite/Cable Broadcast, etc. In these cases, the payload of TS packets contain packetized elementary stream (PES) packets. Further, digital video discs (DVDs) and Blu-Ray Discs (BDs) utilize PES streams where the payload of each PES packet is encrypted.
In operation, video codec 103 encodes, decodes or transcodes the video signal 110 into a processed video signal 112. The pattern recognition module 125 operates cooperatively with the video codec 103, in parallel or in tandem, and optionally based on feedback data from the video codec 103 generated in conjunction with the encoding, decoding or transcoding of the video signal 110. The pattern recognition module 125 processes image sequences in the video signal 110 to detect patterns of interest that, for example, indicate the content of the video signal 110 and the processed video signal 112. When one or more patterns of interest are detected, the pattern recognition module 125 generates pattern recognition data, in response, that indicates the pattern or patterns of interest. The pattern recognition data can take the form of data that identifies patterns and corresponding features, like color, shape, size information, number and motion, the recognition of objects or features, as well as the location of these patterns or features in regions of particular images of an image sequence as well as the addresses, time stamps or other identifiers of the images in the sequence that contain these particular objects or features.
In addition to color histogram data, other coding feedback generated by the video codec 103 in the video encoding/decoding or transcoding can be employed to aid the process of recognizing the content in the processed video signal 112. For example, while temporal and spatial information is used by video codec 103 to remove redundancy, this information can also be used by pattern recognition module 125 to detect or recognize features like sky, grass, sea, wall, buildings and building features such as the type of building, the number of building stories, etc., moving vehicles and animals (including people). Temporal feedback in the form of motion vectors estimated in encoding or retrieved in decoding (or motion information gotten by optical flow for very low resolution) can be used by pattern recognition module 125 for motion-based pattern partition or recognition via a variety of moving group algorithms. In addition, temporal information can be used by pattern recognition module 125 to improve recognition by temporal noise filtering, providing multiple picture candidates to be selected from for recognition of the best image in an image sequence, as well as for recognition of temporal features over a sequence of images. Spatial information such as statistical information, like variance, frequency components and bit consumption estimated from input YUV or retrieved for input streams, can be used for texture based pattern partition and recognition by a variety of different classifiers. More recognition features, like structure, texture, color and motion characters can be used for precise pattern partition and recognition. For instance, line structures can be used to identify and characterize manmade objects such as building and vehicles. Random motion, rigid motion and relative position motion are effective to discriminate water, vehicles and animal respectively. Shot transition information that identifies temporal segments in the image sequence corresponding to a plurality of video shots, group of picture structure, scene transitions and/or other temporal information from encoding or decoding that identifies transitions between video shots in an image sequence can be used to delineate index data 115 into segments and/or to start new pattern detecting and reorganization and provide points of demarcation for temporal recognition across a plurality of images.
In addition, feedback from the pattern recognition module 125 can be used to guide the encoding or transcoding performed by video codec 103. After pattern recognition, more specific structural and statistically information can be retrieved that can guide mode decision and rate control to improve quality and performance in encoding or transcoding of the video signal 110. Pattern recognition can also generate feedback that identifies regions with different characteristics. These more contextually correct and grouped motion vectors can improve quality and save bits for encoding, especially in low bit rate cases. After pattern recognition, estimated motion vectors can be grouped and processed in accordance with the feedback. In particular, pattern recognition feedback can be used by video codec 103 for bit allocation in different regions of an image or image sequence in encoding or transcoding of the video signal 110. With pattern recognition and the codec running together, they can provide powerful aids to each other.
In conjunction with the encoding, decoding and/or transcoding of the video signal 110, the video codec 103 generates or retrieves the decoded image sequence of the content of video signal 110 along with coding feedback for transfer to the pattern recognition module 125. The pattern recognition module 125 operate based on an image sequence to generate pattern recognition data and index data 115 and optionally pattern recognition feedback for transfer back the video codec 103. In particular, pattern recognition module 125 can operate via clustering, statistical pattern recognition, syntactic pattern recognition or via other pattern detection algorithms or methodologies to detect a pattern of interest in an image or image sequence (frame or field) of video signal 110 and generate pattern recognition data and index data 115 in response thereto. The pattern recognition module 125 generates index data 115 to delineate a plurality of segments of the processed video signal 112 and to identify or characterize the content in each segment. The index data 115 can be output via the signal interface 198 in association with the processed video signal 112. While shown as separate signals index data 115 can be provided as metadata to the processed video signal 112 and incorporated in the signal itself as a watermark, video blanking signal or as other data within the processed video signal 112.
The processing module 230 can be implemented using a single processing device or a plurality of processing devices. Such a processing device may be a microprocessor, co-processors, a micro-controller, digital signal processor, microcomputer, central processing unit, field programmable gate array, programmable logic device, state machine, logic circuitry, analog circuitry, digital circuitry, and/or any device that manipulates signals (analog and/or digital) based on operational instructions that are stored in a memory, such as memory module 232. Memory module 232 may be a single memory device or a plurality of memory devices. Such a memory device can include a hard disk drive or other disk drive, read-only memory, random access memory, volatile memory, non-volatile memory, static memory, dynamic memory, flash memory, cache memory, and/or any device that stores digital information. Note that when the processing module implements one or more of its functions via a state machine, analog circuitry, digital circuitry, and/or logic circuitry, the memory storing the corresponding operational instructions may be embedded within, or external to, the circuitry comprising the state machine, analog circuitry, digital circuitry, and/or logic circuitry.
Processing module 230 and memory module 232 are coupled, via bus 250, to the signal interface 198 and a plurality of other modules, such as pattern recognition module 125, decoder section 240 and encoder section 236. In an embodiment of the present disclosure, the signal interface 198, video codec 103 and pattern recognition module 125 each operate in conjunction with the processing module 230 and memory module 232. The modules of video processing system 102 can each be implemented in software, firmware or hardware, depending on the particular implementation of processing module 230. It should also be noted that the software implementations of the present disclosure can be stored on a tangible storage medium such as a magnetic or optical disk, read-only memory or random access memory and also be produced as an article of manufacture. While a particular bus architecture is shown, alternative architectures using direct connectivity between one or more modules and/or additional busses can likewise be implemented in accordance with the present disclosure.
The pattern recognition module 125 includes a shot segmentation module 150 that segments the image sequence 310 into shot data 154 corresponding to the plurality of shots, scenes or other segments, based on the coding feedback data 300. A pattern detection module 175 analyzes the shot data 154 and generates pattern recognition data 156 that identifies content in conjunction with at least one of the plurality of shots, based audio data 312, and further based on color histogram data and optionally other spatial and temporal coding data included in the coding feedback data 300.
In an embodiment, the shot segmentation module 150 operates based on coding feedback data 300 that includes shot transition data 152 generated, for example, by preprocessing information, like variance and downscaled motion cost in encoding; and based on reference and bit consumption information in decoding. Shot transition data 152 can not only be included in coding feedback data 300, but also generated by video codec 103 for use in GOP structure decision, mode selection and rate control to improve quality and performance in encoding.
For example, encoding preprocessing information, like variance and downscaled motion cost, can be used for shot segmentation. Based on their historical tracks, if variance and downscaled motion cost change dramatically, an abrupt shot transitions happens; when variances keep changing monotonously and motion costs jump up and down at the start and end points of the monotonous variance changes, there is a gradual shot transition, like fade-in, fade-out, dissolve, and wipe. In decoding, frame reference information and bit consumption can be used similarly. The output shot transition data 152 can be used not only for GOP structure decision, mode selection and rate control to improve quality and performance in encoding, but also for temporal segmentation of the image sequence 310 and as an enabler for frame-rate invariant shot level searching features.
Index data 115 can include one or more text strings or other identifiers that indicate patterns of interest and other content for use in characterizing segments of the video signal. In addition to video navigation, the index data 115 and be used in video storage and retrieval, and particularly to find videos of interest (e.g. relating to sports or cooking), locate videos containing certain scenes (e.g. a man and a woman on a beach), certain subject matter (e.g. regarding the American Civil War), certain places or venues (e.g. the Eiffel Tower) certain objects (e.g. a Patek Phillipe watch), certain themes (e.g. romance, action, horror), etc. Video indexing can be subdivided into five steps: modeling based on domain-specific attributes, segmentation, extraction, representation, organization. Some functions, like shot (temporally and visually connected frames) and scene (temporally and contextually connected shots) segmentation, used in encoding can likewise be used in visual indexing.
In operation, the pattern detection module 175 operates via clustering, statistical pattern recognition, syntactic pattern recognition or via other pattern detection algorithms or methodologies to detect a pattern of interest in an image or image sequence 310 and generates pattern recognition data 156 in response thereto. In this fashion, object/features in each shot can be correlated to the shots that contain these objects and features that can be used for indexing and search of indexed video for key objects/features and the shots that contain these objects/features. The index data 115 can be used for scene segmentation in a server, set-top box or other video processing system based on the extracted information and algorithms such as a hidden Markov model (HMM) algorithm that is based on a priori field knowledge.
Consider an example where video signal 110 contains a video broadcast. Index data 115 that indicates anchor shots and field shots shown alternately could indicate a news broadcast; crowd shots and sports shots shown alternately could indicate a sporting event. Scene information can also be used for rate control, like quantization parameter (QP) initialization at shot transition in encoding. Index data 115 can be used to generate more high-level motive and contextual descriptions via manual review by human personnel. For instance, based on results mentioned above, operators could process index data 115 to provide additional descriptors for an image sequence 310 to, for example, describe an image sequence as “around 10 people (Adam, Brian . . . ) watching a live Elton John show on grass under the sky in the Queen's Park, where Elton John is performing Rocket Man.”
The index data 115 can contain pattern recognition data 156 and other hierarchical indexing information like: frame-level temporal and spatial information including variance, global motion and bit number etc.; shot-level objects and text string or other descriptions of features such as text regions of a video, human and action description, object information and background texture description etc.; scene-level representations such as video category (news cast, sitcom, commercials, movie, sports or documentary etc.), and high-level context-level descriptions and presentations presented as text strings, numerical classifiers or other data descriptors.
In addition, pattern recognition feedback 298 in the form of pattern recognition data 156 or other feedback from the pattern recognition module 125 can be used to guide the encoding or transcoding performed by video codec 103. After pattern recognition, more specific structural and statistically information can be generated as pattern recognition feedback 298 that can, for instance, guide mode decision and rate control to improve quality and performance in encoding or transcoding of the video signal 110. Pattern recognition module 125 can also generate pattern recognition feedback 298 that identifies regions with different characteristics. These more contextually correct and grouped motion vectors can improve quality and save bits for encoding, especially in low bit rate cases. After pattern recognition, estimated motion vectors can be grouped and processed in accordance with the pattern recognition feedback 298. In particular, the pattern recognition feedback 298 can be used by video codec 103 for bit allocation in different regions of an image or image sequence in encoding or transcoding of the video signal 110.
The coding feedback data 300 can be generated by video codec 103 in conjunction with either a decoding of the video signal 110, an encoding of the video signal 110 or a transcoding of the video signal 110. The video codec 103 can generate the shot transition data 152 based on image statistics, group of picture data, etc. As discussed above, encoding preprocessing information, like variance and downscaled motion cost, can be used to generate shot transition data 152 for shot segmentation. Based on their historical tracks, if variance and downscaled motion cost change dramatically, an abrupt shot transitions happens; when variances keep changing monotonously and motion costs jump up and down at the start and end points of the monotonous variance changes, there is a gradual shot transition, like fade-in, fade-out, dissolve, and wipe. In decoding, frame reference information and bit consumption can be used similarly. The output shot transition data 152 can be used not only for GOP structure decisions, mode selection and rate control to improve quality and performance in encoding, but also for temporal segmentation of the image sequence 310 and as an enabler for frame-rate invariant shot level searching features. While the foregoing has focused on the delineation of shots based on purely video and image data, associated audio data can be used in addition to or in the alternative to video data as a way of delineating and characterizing video segments. For example, one or more shots of a video programs can be delineated based the start and stop of a song, other distinct audio sounds, such as running water, wind or other storm sounds or other audio content of a sound track corresponding to the video signal.
Further, coding feedback data 300 and audio data 312 can also be used by pattern detection module 175. The pattern recognition module 125 can generate the pattern recognition data 156 based on audio data 312, coding feedback data 300 that includes color histogram data and optionally one or more other image statistics to identify features such as faces, text, places, music, human actions, as well as other objects and features. As previously discussed, temporal and spatial information used by video codec 103 to remove redundancy can also be used by pattern detection module 175 to detect or recognize features like sky, grass, sea, wall, buildings, moving vehicles and animals (including people). Temporal feedback in the form of motion vectors estimated in encoding or retrieved in decoding (or motion information gotten by optical flow for very low resolution) can be used by pattern detection module 175 for motion-based pattern partition or recognition via a variety of moving group algorithms. Spatial information such as statistical information, like variance, frequency components and bit consumption estimated from input YUV or retrieved for input streams, can be used for texture based pattern partition and recognition by a variety of different classifiers. More recognition features, like structure, texture, color and motion characters can be used for precise pattern partition and recognition. For instance, line structures can be used to identify and characterize manmade objects such as building and vehicles. Random motion, rigid motion and relative position motion are effective to discriminate water, vehicles and animal respectively.
In addition to analysis of static images included in the shot data 154, shot data 154 can includes a plurality of images in the image sequence 310, and the pattern detection module 175 can generate the pattern recognition data 156 based on a temporal recognition performed over a plurality of images within a shot. Slight motion within a shot and aggregation of images over a plurality of shots can enhance the resolution of the images for pattern analysis, can provide three-dimensional data from differing perspectives for the analysis and recognition of three-dimensional objects and other motion can aid in recognizing objects and other features based on the motion that is detected.
Pattern detection module 175 generates the pattern feedback data 298 as described in conjunction with
In an example of operation, the video processing system 102 is part of a web server, teleconferencing system security system or set top box that generates index data 115 with facial recognition. The pattern detection module 175 operates based on coding feedback data 300 that include motion vectors estimated in encoding or retrieved in decoding (or motion information gotten by optical flow etc. for very low resolution), together with a skin color model used to roughly partition face candidates. The pattern detection module 175 tracks a candidate facial region over the plurality of images and detects a face in the image based on the one or more of these images. Shot transition data 152 in coding feedback data 300 can be used to start a new series of face detecting and tracking.
For example, pattern detection module 175 can operate via color histogram data to detect colors in image sequence 310. The pattern detection module 175 generates a color bias corrected image from image sequence 310 and a color transformed image from the color bias corrected image. Pattern detection module 175 then operates to detect colors in the color transformed image that correspond to skin tones. In particular, pattern detection module 175 can operate using an elliptic skin model in the transformed space such as a CbCr subspace of a transformed YCbCr space. In particular, a parametric ellipse corresponding to contours of constant Mahalanobis distance can be constructed under the assumption of Gaussian skin tone distribution to identify a detected region 322 based on a two-dimension projection in the CbCr subspace. As exemplars, the 853,571 pixels corresponding to skin patches from the Heinrich-Hertz-Institute image database can be used for this purpose, however, other exemplars can likewise be used in broader scope of the present disclosure.
In an embodiment, the pattern detection module 175 tracks a candidate facial region over the plurality of images and detects a facial region based on an identification of facial motion in the candidate facial region over the plurality of images, wherein the facial motion includes at least one of: eye movement; and the mouth movement. In particular, face candidates can be validated for face detection based on the further recognition by pattern detection module 175 of facial features, like eye blinking (both eyes blink together, which discriminates face motion from others; the eyes are symmetrically positioned with a fixed separation, which provides a means to normalize the size and orientation of the head), shape, size, motion and relative position of face, eyebrows, eyes, nose, mouth, cheekbones and jaw. Any of these facial features can be used extracted from the shot data 154 and used by pattern detection module 175 to eliminate false detections. Further, the pattern detection module 175 can employ temporal recognition to extract three-dimensional features based on different facial perspectives included in the plurality of images to improve the accuracy of the recognition of the face. Using temporal information, the problems of face detection including poor lighting, partially covering, size and posture sensitivity can be partly solved based on such facial tracking. Furthermore, based on profile view from a range of viewing angles, more accurate and 3D features such as contour of eye sockets, nose and chin can be extracted.
In addition to generating pattern recognition data 156 for indexing, the pattern recognition data 156 that indicates a face has been detected and the location of the facial region can also be used as pattern recognition feedback 298. The pattern recognition data 156 can include facial characteristic data such as position in stream, shape, size and relative position of face, eyebrows, eyes, nose, mouth, cheekbones and jaw, skin texture and visual details of the skin (lines, patterns, and spots apparent in a person's skin), or even enhanced, normalized and compressed face images. In response, the encoder section 236 can guide the encoding of the image sequence based on the location of the facial region. In addition, pattern recognition feedback 298 that includes facial information can be used to guide mode selection and bit allocation during encoding. Further, the pattern recognition data 156 and pattern recognition feedback 298 can further indicate the location of eyes or mouth in the facial region for use by the encoder section 236 to allocate greater resolution to these important facial features. For example, in very low bit rate cases the encoder section 236 can avoid the use of inter-mode coding in the region around blinking eyes and/or a talking mouth, allocating more encoding bits should to these face areas.
In a further example of operation, the video processing system 102 is part of a web server, teleconferencing system security system or set top box that generates index data 115 with text recognition. In this fashion, text data such as automobile license plate numbers, store signs, building names, subtitles, name tags, and other text portions in the image sequence 310 can be detected and recognized. Text regions typically have obvious features that can aid detection and recognition. These regions have relatively high frequency; they are usually have high contrast in a regular shape; they are usually aligned and spaced equally; they tend to move with background or objects.
Coding feedback data 300 can be used by the pattern detection module 175 to aid in detection. For example, shot transition data from encoding or decoding can be used to start a new series of text detecting and tracking Statistical information, like variance, frequency component and bit consumption, estimated from input YUV or retrieved from input streams can be used for text partitioning. Edge detection, YUV projection, alignment and spacing information, etc. can also be used to further partition interest text regions. Coding feedback data 300 in the form of motion vectors can be retrieved for the identified text regions in motion compensation. Then reliable structural features, like lines, ends, singular points, shape and connectivity can be extracted.
In this mode of operation, the pattern detection module 175 generates pattern recognition data 156 that can include an indication that text was detected, a location of the region of text and index data 115 that correlates the region of text to a corresponding video shots. The pattern detection module 175 can further operate to generate a text string by recognizing the text in the region of text and further to generate index data 115 that includes the text string correlated to the corresponding video shot. The pattern recognition module 125 can operate via a trained hierarchical and fuzzy classifier, neural network and/or vector processing engine to recognize text in a text region and to generate candidate text strings. These candidate text strings may optionally be modified later into final text by post processing or further offline analysis and processing of the shot data.
The pattern recognition data 156 can be included in pattern recognition feedback 298 and used by the encoder section 236 to guide the encoding of the image sequence. In this fashion, text region information can guide mode selection and rate control. For instance, small partition mode can be avoided in a small text region; motions vector can be grouped around text; and high quantization steps can be avoided in text regions, even in very low bit rate case to maintain adequate reproduction of the text.
In another example of operation, the video processing system 102 is part of a web server, teleconferencing system security system or set top box that generates index data 115 with recognition of human action. In this fashion and region of human action can be determined along with the determination of human action descriptions such as a number of people, body sizes and features, pose types, position, velocity and actions such as kick, throw, catch, run, walk, fall down, loiter, drop an item, etc. can be detected and recognized.
Coding feedback data 300 can be used by the pattern detection module 175 to aid in detection and tracking of events and actions. For example, shot transition data from encoding or decoding can be used to start a new series of action detecting and tracking Motion vectors from encoding or decoding (or motion information gotten by optical flow etc. for very low resolution) can be employed for this purpose.
In this mode of operation, the pattern detection module 175 generates pattern recognition data 156 that can include an indication that human was detected, a location of the region of the human and index data 115 that includes, for example human action descriptors and correlates the human action to a corresponding video shot. The pattern detection module 175 can subdivide the process of human action recognition into: moving object detecting, human discriminating, tracking, action understanding and recognition. In particular, the pattern detection module 175 can identify a plurality of moving objects in the plurality of images. For example, motion objects can be partitioned from background. The pattern detection module 175 can then discriminate one or more humans from the plurality of moving objects. Human motion can be non-rigid and periodic. Shape-based features, including color and shape of face and head, width-height-ratio, limb positions and areas, tile angle of human body, distance between feet, projection and contour character, etc. can be employed to aid in this discrimination. These shape, color and/or motion features can be recognized as corresponding to human action via a classifier such as neural network. The action of the human can be tracked over the images in a shot and a particular type of human action can be recognized in the plurality of images. Individuals, presented as a group of corners and edges etc., can be precisely tracked using algorithms such as model-based and active contour-based algorithm. Gross moving information can be achieved via a Kalman filter or other filter techniques. Based on the tracking information, action recognition can be implemented by Hidden Markov Model, dynamic Bayesian networks, syntactic approaches or via other pattern recognition algorithm.
In an embodiment, the pattern detection module operates based on a classifier function that maps an input attribute vector, x=(x1, x2, x3, x4, . . . , xn), to a confidence that the input belongs to a class, that is, f(x)=confidence(class). The input attribute data can include a color histogram data, audio data, image statistics, motion vector data, other coding feedback data 300 and other attributes extracted from the image sequence 310. Such classification can employ a probabilistic and/or statistical-based analysis (e.g., factoring into the analysis utilities and costs) to prognose or infer an action that a user desires to be automatically performed. A support vector machine (SVM) is an example of a classifier that can be employed. The SVM operates by finding a hypersurface in the space of possible inputs, which the hypersurface attempts to split the triggering criteria from the non-triggering events. This makes the classification correct for testing data that is near, but not identical to training data. Other directed and undirected model classification approaches comprise, e.g., naïve Bayes, Bayesian networks, decision trees, neural networks, fuzzy logic models, and probabilistic classification models providing different patterns of independence can be employed. Classification as used herein also is inclusive of statistical regression that is utilized to develop models of priority.
As will be readily appreciated, one or more of the embodiments can employ classifiers that are explicitly trained (e.g., via a generic training data) as well as implicitly trained (e.g., via observing UE behavior, operator preferences, historical information, receiving extrinsic information). For example, SVMs can be configured via a learning or training phase within a classifier constructor and feature selection module.
It should be noted that classifiers functions containing multiple different kinds of attribute data can provide a powerful approach to recognition. In one mode of operation, the pattern detection module 175 can recognize content that includes an object, based on color histogram data corresponding to colors of the object and sound data corresponding to a sound of the object and optionally other features. For example, a Coke bottle or can be recognized based on a distinctive color histogram, a shape corresponding to a bottle or can, the sound of a bottle or can being opened, and further based on text recognition of a Coca-Cola label.
In another mode of operation, the pattern detection module 175 can recognize content that includes a human activity, based on color histogram data and sound data corresponding to a sound of the activity and optionally other features. For example, a kick-off in a football game can be recognized based on color histogram data corresponding to a team's uniforms and a particular region that includes colors corresponding to a football, and further based on the sound of a football being kicked.
In another mode of operation, the pattern detection module 175 can recognize content that includes a person, based on color histogram data corresponding to colors of the person's face and sound data corresponding to a voice of the person. For example, color histogram data can be used to identify a region that contains a face, facial and speaker recognition can be used together to identify an actor in a scene as Brad Pitt.
In another mode of operation, the pattern detection module 175 can recognize content that includes a place, based on color histogram data corresponding to colors of the place, image data corresponding to a recognized shape and sound data corresponding to a sound of the place. For example, the Niagara Falls can be recognized based on scene motion or texture data, a color histogram corresponding to rushing water and sound data corresponding to the sound of the falls.
The pattern detection module 175 generates index data 115 that includes pattern recognition data 156 in conjunction with each of the shots that identifies the first shot as being a commentator shot, the second and fourth shots as being shots of the game and the third shot is being a shot of the crowd. The pattern recognition data 156 is correlated to the shot transition data 152 to generating index data 115 that identifies the location of each shot in the image sequence 310 and to associate each shot with the corresponding pattern recognition data 156, an optionally to identify a region within the shot by image and/or within one or more images that include the identified subject matter.
In an embodiment, the pattern recognition module 125 identifies a football in the scene, the teams that are playing in the game based on analysis of the color and images associated with their uniforms and based on text data contained in the video program. The pattern recognition module 125 can further identify which team has the ball (the team in possession) not only to generate index data 115 that characterizes various game shots as plays, but further to characterize the team that is running the play, but also the type of play, a pass, a run, a turnover, a play where player X has the ball, a scoring play that results in a touchdown or field goal, a punt or kickoff, plays that excited the crowd in the stadium, players that were the subject of official review, etc.
In the example shown, a first play of the game (Play #1) can contain the kickoff by the away team to the home team. This first play is followed by inter-play activity such as a crowd shot. The inter-play activity is followed by Play#2, such as the opening play of the drive by the home team. The index data 115 can not only identify an address range that delineates each of these three segments of the video but also includes characteristics that define each segment as being either a play or inter-play activity but optionally includes further characteristics that further characterize or define each play and the inter-play activity.
As discussed in conjunction with
While the foregoing has focused on one type of index data 115 for a particular type of content, i.e. a football game, the processing system 102 can operate to generate index data 115 of different kinds for different sporting events, for different events and for different types of video content such as documentaries, motion pictures, news broadcasts, video clips, infomercials, reality television programs and other television shows, and other content.
The generation of index data 115 in this fashion allows a user to navigate video content in a processed video signal 112 in a non-linear (.e. not in linear or temporal order), non-contiguous, multilayer and/or other non-traditional fashion. Consider an example where the user of a video player has downloaded this football game and the associated index data 115. The user could choose to watch only plays of the home team—in effect, viewing the game in a non-contiguous fashion, skipping over other portions of the game. The user could also view the game out of temporal order by first watching only the scoring plays of the game. If the game seems to be of more interest, the user could change chapter modes to start back from the beginning and watching all of the plays of the game for each team.
As discussed in conjunction with
Considering, for example, the case where the shot data 154 includes a human face and the pattern detection module 175 generates a region corresponding the human face, candidate region detection module 320 can generate detected region 322 based on the detection of pixel color values corresponding to facial features such as skin tones. Region cleaning module can generate a more contiguous region that contains these facial features and region growing module can grow this region to include the surrounding hair and other image portions to ensure that the entire face is included in the region identified by region identification data 330.
As previously discussed, the encoder feedback data 296 includes shot transition data, such as shot transition data 152, that identifies temporal segments in the image sequence 310 that are used to bound the shot data 154 to a particular set of images in the image sequence 310. The candidate region detection module 320 further operates based on motion vector data to track the position of candidate region through the images in the shot data 154. Motion vectors, shot transition data and other encoder feedback data 296 are also made available to region tracking and accumulation module 334 and region recognition module 350. The region tracking and accumulation module 334 provides accumulated region data 336 that includes a temporal accumulation of the candidate regions of interest to enable temporal recognition via region recognition module 350. In this fashion, region recognition module 350 can generate pattern recognition data based on such features as facial motion, human actions, three-dimensional modeling and other features recognized and extracted based on such temporal recognition.
The supplemental pattern recognition module 360 can be implemented using a single processing device or a plurality of processing devices. Such a processing device may be a microprocessor, co-processors, a micro-controller, digital signal processor, microcomputer, central processing unit, field programmable gate array, programmable logic device, state machine, logic circuitry, analog circuitry, digital circuitry, and/or any device that manipulates signals (analog and/or digital) based on operational instructions that are stored in a memory. Such a memory may be a single memory device or a plurality of memory devices. Such a memory device can include a hard disk drive or other disk drive, read-only memory, random access memory, volatile memory, non-volatile memory, static memory, dynamic memory, flash memory, cache memory, and/or any device that stores digital information. Note that when the supplemental pattern recognition module 360 implements one or more of its functions via a state machine, analog circuitry, digital circuitry, and/or logic circuitry, the memory storing the corresponding operational instructions may be embedded within, or external to, the circuitry comprising the state machine, analog circuitry, digital circuitry, and/or logic circuitry.
A previously discussed, the index data generated in this fashion could be used to generate a searchable index of this video along with other video as part of a video search system. A user of the video processing system 102 could search videos for “Sheffield Wednesday” and not only identify the particular video broadcast, but also identify the particular shot or shots within the video, such as the shot containing image 370, that contain a text region, such as text region 372 that generated the search string “Sheffield Wednesday”.
For instance, following with the example discussed in conjunction with
The region recognition module 350 or supplemental pattern recognition module 360 operate based on accumulated regions 391, 393 and 395 that include similar regions containing the punt to generate further pattern recognition data 156 that includes human action descriptors such as “football play”, “kick”, “punt”, information regarding the distance, height, trajectory of the ball and/or other descriptors that characterize this particular action.
It should be noted, that while the descriptions of
The transmission path 122 can include a wireless path that operates in accordance with a wireless local area network protocol such as an 802.11 protocol, a WIMAX protocol, a Bluetooth protocol, etc. Further, the transmission path can include a wired path that operates in accordance with a wired protocol such as a Universal Serial Bus protocol, an Ethernet protocol or other high speed protocol.
While these particular devices are illustrated, video storage system 79 can include a hard drive, flash memory device, computer, DVD burner, or any other device that is capable of generating, storing, encoding, decoding, transcoding and/or displaying a video signal in accordance with the methods and systems described in conjunction with the features and functions of the present disclosure as described herein.
In an embodiment, the coding feedback data further includes shot transition data that identifies temporal segments in the image sequence corresponding to a plurality of video shots. The shots can include a plurality of images in the image sequence and the index data can be generated based on a temporal recognition performed over the plurality of images. Step 400 can include recognizing content that includes an object, based on color histogram data corresponding to colors of the object and sound data corresponding to a sound of the object. Step 400 can include recognizing content that includes a person, based on color histogram data corresponding to colors of the person's face and sound data corresponding to a voice of the person. Step 400 can include recognizing content that includes a human activity, based on color histogram data corresponding to colors of the human activity and sound data corresponding to a sound of the human activity. Step 400 can include recognizing content further based on a recognized shape. Step 400 can include recognizing content that includes a place, based on color histogram data corresponding to colors of the place, image data corresponding to a recognized shape and sound data corresponding to a sound of the place.
It is noted that terminologies as may be used herein such as bit stream, stream, signal sequence, etc. (or their equivalents) have been used interchangeably to describe digital information whose content corresponds to any of a number of desired types (e.g., data, video, speech, audio, etc. any of which may generally be referred to as ‘data’).
As may be used herein, the terms “substantially” and “approximately” provides an industry-accepted tolerance for its corresponding term and/or relativity between items. Such an industry-accepted tolerance ranges from less than one percent to fifty percent and corresponds to, but is not limited to, component values, integrated circuit process variations, temperature variations, rise and fall times, and/or thermal noise. Such relativity between items ranges from a difference of a few percent to magnitude differences. As may also be used herein, the term(s) “configured to”, “operably coupled to”, “coupled to”, and/or “coupling” includes direct coupling between items and/or indirect coupling between items via an intervening item (e.g., an item includes, but is not limited to, a component, an element, a circuit, and/or a module) where, for an example of indirect coupling, the intervening item does not modify the information of a signal but may adjust its current level, voltage level, and/or power level. As may further be used herein, inferred coupling (i.e., where one element is coupled to another element by inference) includes direct and indirect coupling between two items in the same manner as “coupled to”. As may even further be used herein, the term “configured to”, “operable to”, “coupled to”, or “operably coupled to” indicates that an item includes one or more of power connections, input(s), output(s), etc., to perform, when activated, one or more its corresponding functions and may further include inferred coupling to one or more other items. As may still further be used herein, the term “associated with”, includes direct and/or indirect coupling of separate items and/or one item being embedded within another item.
As may be used herein, the term “compares favorably”, indicates that a comparison between two or more items, signals, etc., provides a desired relationship. For example, when the desired relationship is that signal 1 has a greater magnitude than signal 2, a favorable comparison may be achieved when the magnitude of signal 1 is greater than that of signal 2 or when the magnitude of signal 2 is less than that of signal 1. As may be used herein, the term “compares unfavorably”, indicates that a comparison between two or more items, signals, etc., fails to provide the desired relationship.
As may also be used herein, the terms “processing module”, “processing circuit”, “processor”, and/or “processing unit” may be a single processing device or a plurality of processing devices. Such a processing device may be a microprocessor, micro-controller, digital signal processor, microcomputer, central processing unit, field programmable gate array, programmable logic device, state machine, logic circuitry, analog circuitry, digital circuitry, and/or any device that manipulates signals (analog and/or digital) based on hard coding of the circuitry and/or operational instructions. The processing module, module, processing circuit, and/or processing unit may be, or further include, memory and/or an integrated memory element, which may be a single memory device, a plurality of memory devices, and/or embedded circuitry of another processing module, module, processing circuit, and/or processing unit. Such a memory device may be a read-only memory, random access memory, volatile memory, non-volatile memory, static memory, dynamic memory, flash memory, cache memory, and/or any device that stores digital information. Note that if the processing module, module, processing circuit, and/or processing unit includes more than one processing device, the processing devices may be centrally located (e.g., directly coupled together via a wired and/or wireless bus structure) or may be distributedly located (e.g., cloud computing via indirect coupling via a local area network and/or a wide area network). Further note that if the processing module, module, processing circuit, and/or processing unit implements one or more of its functions via a state machine, analog circuitry, digital circuitry, and/or logic circuitry, the memory and/or memory element storing the corresponding operational instructions may be embedded within, or external to, the circuitry comprising the state machine, analog circuitry, digital circuitry, and/or logic circuitry. Still further note that, the memory element may store, and the processing module, module, processing circuit, and/or processing unit executes, hard coded and/or operational instructions corresponding to at least some of the steps and/or functions illustrated in one or more of the Figures. Such a memory device or memory element can be included in an article of manufacture.
One or more embodiments have been described above with the aid of method steps illustrating the performance of specified functions and relationships thereof. The boundaries and sequence of these functional building blocks and method steps have been arbitrarily defined herein for convenience of description. Alternate boundaries and sequences can be defined so long as the specified functions and relationships are appropriately performed. Any such alternate boundaries or sequences are thus within the scope and spirit of the claims. Further, the boundaries of these functional building blocks have been arbitrarily defined for convenience of description. Alternate boundaries could be defined as long as the certain significant functions are appropriately performed. Similarly, flow diagram blocks may also have been arbitrarily defined herein to illustrate certain significant functionality.
To the extent used, the flow diagram block boundaries and sequence could have been defined otherwise and still perform the certain significant functionality. Such alternate definitions of both functional building blocks and flow diagram blocks and sequences are thus within the scope and spirit of the claims. One of average skill in the art will also recognize that the functional building blocks, and other illustrative blocks, modules and components herein, can be implemented as illustrated or by discrete components, application specific integrated circuits, processors executing appropriate software and the like or any combination thereof.
In addition, a flow diagram may include a “start” and/or “continue” indication. The “start” and “continue” indications reflect that the steps presented can optionally be incorporated in or otherwise used in conjunction with other routines. In this context, “start” indicates the beginning of the first step presented and may be preceded by other activities not specifically shown. Further, the “continue” indication reflects that the steps presented may be performed multiple times and/or may be succeeded by other activities not specifically shown. Further, while a flow diagram indicates a particular ordering of steps, other orderings are likewise possible provided that the principles of causality are maintained.
The one or more embodiments are used herein to illustrate one or more aspects, one or more features, one or more concepts, and/or one or more examples. A physical embodiment of an apparatus, an article of manufacture, a machine, and/or of a process may include one or more of the aspects, features, concepts, examples, etc. described with reference to one or more of the embodiments discussed herein. Further, from figure to figure, the embodiments may incorporate the same or similarly named functions, steps, modules, etc. that may use the same or different reference numbers and, as such, the functions, steps, modules, etc. may be the same or similar functions, steps, modules, etc. or different ones.
Unless specifically stated to the contra, signals to, from, and/or between elements in a figure of any of the figures presented herein may be analog or digital, continuous time or discrete time, and single-ended or differential. For instance, if a signal path is shown as a single-ended path, it also represents a differential signal path. Similarly, if a signal path is shown as a differential path, it also represents a single-ended signal path. While one or more particular architectures are described herein, other architectures can likewise be implemented that use one or more data buses not expressly shown, direct connectivity between elements, and/or indirect coupling between other elements as recognized by one of average skill in the art.
The term “module” is used in the description of one or more of the embodiments. A module implements one or more functions via a device such as a processor or other processing device or other hardware that may include or operate in association with a memory that stores operational instructions. A module may operate independently and/or in conjunction with software and/or firmware. As also used herein, a module may contain one or more sub-modules, each of which may be one or more modules.
While particular combinations of various functions and features of the one or more embodiments have been expressly described herein, other combinations of these features and functions are likewise possible. The present disclosure is not limited by the particular examples disclosed herein and expressly incorporates these other combinations.
The present application claims priority under 35 U.S.C. 120 as a continuation-in-part of the U.S. Application entitled, VIDEO PROCESSING SYSTEM WITH PATTERN DETECTION AND METHODS FOR USE THEREWITH, having Ser. No. 13/467,522, and filed on May 9, 2012, that itself claims priority under 35 USC 119(e) to the provisionally filed U.S. Application entitled, VIDEO PROCESSING SYSTEM WITH PATTERN DETECTION AND METHODS FOR USE THEREWITH, having Ser. No. 61/635,034, and filed on Apr. 18, 2012, the contents of which are expressly incorporated herein in their entirety by reference for any and all purposes.
| Number | Date | Country | |
|---|---|---|---|
| 61635034 | Apr 2012 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 13467522 | May 2012 | US |
| Child | 14556887 | US |
