EP2815572A2 - Encodage et décodage vidéo basés sur le contexte - Google Patents

Encodage et décodage vidéo basés sur le contexte

Info

Publication number
EP2815572A2
EP2815572A2 EP13710636.5A EP13710636A EP2815572A2 EP 2815572 A2 EP2815572 A2 EP 2815572A2 EP 13710636 A EP13710636 A EP 13710636A EP 2815572 A2 EP2815572 A2 EP 2815572A2
Authority
EP
European Patent Office
Prior art keywords
feature
macroblock
prediction
encoding
frames
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Withdrawn
Application number
EP13710636.5A
Other languages
German (de)
English (en)
Inventor
Nigel Lee
Renato Pizzorni
Darin Deforest
Charles P. Pace
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Euclid Discoveries LLC
Original Assignee
Euclid Discoveries LLC
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority claimed from US13/725,940 external-priority patent/US9578345B2/en
Application filed by Euclid Discoveries LLC filed Critical Euclid Discoveries LLC
Publication of EP2815572A2 publication Critical patent/EP2815572A2/fr
Withdrawn legal-status Critical Current

Links

Classifications

    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/85Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using pre-processing or post-processing specially adapted for video compression
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/20Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using video object coding
    • H04N19/23Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using video object coding with coding of regions that are present throughout a whole video segment, e.g. sprites, background or mosaic
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/50Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding
    • H04N19/503Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding involving temporal prediction
    • H04N19/51Motion estimation or motion compensation

Definitions

  • Video compression can be considered the process of representing digital video data in a form that uses fewer bits when stored or transmitted.
  • Video compression algorithms can achieve compression by exploiting redundancies and irrelevancies in the video data, whether spatial, temporal, or color-space.
  • Video compression algorithms typically segment the video data into portions, such as groups of frames and groups of pels, to identify areas of redundancy within the video that can be represented with fewer bits than the original video data. When these redundancies in the data are reduced, greater compression can be achieved.
  • An encoder can be used to transform the video data into an encoded format, while a decoder can be used to transform encoded video back into a form comparable to the original video data.
  • the implementation of the encoder/decoder is referred to as a codec.
  • Standard encoders divide a given video frame into non-overlapping coding units or macroblocks (rectangular regions of contiguous pels) for encoding.
  • the macroblocks are typically processed in a traversal order of left to right and top to bottom in the frame. Compression can be achieved when macroblocks are predicted and encoded using previously-coded data.
  • the process of encoding macroblocks using spatially neighboring samples of previously-coded macroblocks within the same frame is referred to as intra-prediction.
  • Intra-prediction attempts to exploit spatial redundancies in the data.
  • the encoding of macroblocks using similar regions from previously-coded frames, together with a motion estimation model is referred to as inter-prediction.
  • Inter-prediction attempts to exploit temporal redundancies in the data.
  • the encoder may measure the difference between the data to be encoded and the prediction to generate a residual.
  • the residual can provide the difference between a predicted macroblock and the original macroblock.
  • the encoder can generate motion vector information that specifies, for example, the location of a macroblock in a reference frame relative to a macroblock that is being encoded or decoded.
  • the predictions, motion vectors (for inter-prediction), residuals, and related data can be combined with other processes such as a spatial transform, a quantizer, an entropy encoder, and a loop filter to create an efficient encoding of the video data.
  • the residual that has been quantized and transformed can be processed and added back to the prediction, assembled into a decoded frame, and stored in a framestore. Details of such encoding techniques for video will be familiar to a person skilled in the art.
  • H.264/MPEG-4 AVC advanced video coding
  • H.264 is a codec standard for video compression that utilizes block-based motion estimation and compensation and achieves high quality video representation at relatively low bitrates.
  • This standard is one of the encoding options used for Blu-ray disc creation and within major video distribution channels, including video streaming on the internet, video conferencing, cable television and direct-broadcast satellite television.
  • the basic coding units for H.264 are 16x16 macroblocks.
  • H.264 is the most recent widely-accepted standard in video compression.
  • the basic MPEG standard defines three types of frames (or pictures), based on how the macroblocks in the frame are encoded.
  • An I-frame (intra-coded picture) is encoded using only data present in the frame itself.
  • the encoder receives video signal data, the encoder creates I frames first and segments the video frame data into macroblocks that are each encoded using intra-prediction.
  • an I- frame consists of only intra-predicted macroblocks (or "intra macroblocks"). I-frames can be costly to encode, as the encoding is done without the benefit of information from previously-decoded frames.
  • a P-frame (predicted picture) is encoded via forward prediction, using data from previously-decoded I-frames or P-frames, also known as reference frames.
  • P-frames can contain either intra macroblocks or (forward-)predicted macroblocks.
  • a B-frame (bi-predictive picture) is encoded via bidirectional prediction, using data from both previous and subsequent frames.
  • B-frames can contain intra, (forward-)predicted, or bi-predicted macroblocks.
  • BBMEC block-based motion estimation and compensation
  • the BBMEC process searches for the best match between the target macroblock (the current macroblock being encoded) and similar- sized regions within previously-decoded reference frames.
  • the encoder may transmit a motion vector.
  • the motion vector may include a pointer to the best match's frame position as well as information regarding the difference between the best match and the corresponding target macroblock.
  • a particular set of reference frames is termed a Group of Pictures (GOP).
  • the GOP contains only the decoded pels within each reference frame and does not include information as to how the macroblocks or frames themselves were originally encoded (I-frame, B-frame or P-frame).
  • Older video compression standards such as MPEG-2, used one reference frame (the previous frame) to predict P-frames and two reference frames (one past, one future) to predict B-frames.
  • the H.264 standard by contrast, allows the use of multiple reference frames for P-frame and B-frame prediction. While the reference frames are typically temporally adjacent to the current frame, there is also accommodation for the specification of reference frames from outside the set of the temporally adjacent frames.
  • the H.264 standard also introduces additional encoding flexibility by dividing frames into spatially distinct regions of one or more contiguous macroblocks called slices. Each slice in a frame is encoded (and can thus be decoded) independently from other slices. I-slices, P-slices, and B-slices are then defined in a manner analogous to the frame types described above, and a frame can consist of multiple slice types.
  • H.264 standard allows a codec to provide better quality video at lower file sizes than previous standards, such as MPEG-2 and MPEG-4 ASP (advanced simple profile)
  • "conventional" compression codecs implementing the H.264 standard typically have struggled to keep up with the demand for greater video quality and resolution on memory-constrained devices, such as smartphones and other mobile devices, operating on limited-bandwidth networks. Video quality and resolution are often compromised to achieve adequate playback on these devices. Further, as video resolution increases, file sizes increase, making storage of videos on and off these devices a potential concern.
  • the present invention recognizes fundamental limitations in the inter-prediction process of conventional codecs and applies higher-level modeling to overcome those limitations and provide improved inter-prediction, while maintaining the same general processing flow and framework as conventional encoders.
  • higher-level modeling provides an efficient way of navigating more of the prediction search space (the video datacube) to produce better predictions than can be found through conventional block-based motion estimation and compensation.
  • computer- vision-based feature and object detection algorithms identify regions of interest throughout the video datacube.
  • the detection algorithm may be from the class of nonparametric feature detection algorithms.
  • the detected features and objects are modeled with a compact set of parameters, and similar feature/object instances are associated across frames.
  • the invention forms tracks out of the associated feature/objects, relates the tracks to specific blocks of video data to be encoded, and uses the tracking information to produce model-based predictions for those blocks of data.
  • the specific blocks of data to be encoded may be macroblocks.
  • the formed tracks relate features to respective macroblocks.
  • Feature/object tracking provides additional context to the conventional encoding/decoding process. Additionally, the modeling of features/objects with a compact set of parameters enables information about the features/objects to be stored efficiently in memory, unlike reference frames, whose totality of pels are expensive to store. Thus, feature/object models can be used to search more of the video datacube, without requiring a prohibitive amount of additional computations or memory. The resulting model-based predictions are superior to conventional inter-predictions, because the model-based predictions are derived from more of the prediction search space.
  • the compact set of parameters includes information about the feature/objects and this set is stored in memory.
  • the respective parameter includes a feature descriptor vector and a location of the feature. The respective parameter is generated when the respective feature is detected.
  • the present invention forms such ensemble matrices, summarizes the matrices using subspaces of important vectors, and uses the vector subspaces as parametric models of the associated features/objects. This can result in especially efficient encodings when those particular features/objects appear in the data.
  • Computer-based methods, codecs and other computer systems and apparatus for processing video data may embody the foregoing principles of the present invention.
  • FIG. 1A is a block diagram depicting feature modeling according to an embodiment of the invention.
  • FIG. IB is a block diagram depicting feature tracking according to an embodiment of the invention.
  • FIG. 1C is a block diagram illustrating the steps in relating features to nearby macrob locks and using the tracks of those features to generate good predictions for those macroblocks, according to an embodiment of the invention.
  • FIG. 2 A is a schematic diagram illustrating the modeling of data at multiple fidelities to provide efficient encodings, according to an embodiment of the invention.
  • FIG. 2B is a block diagram illustrating the identification of objects through feature model correlation and aggregation, according to an embodiment of the invention.
  • FIG. 2C is a block diagram illustrating the identification of objects via aggregation of both nearby features and nearby macroblocks, according to an embodiment of the invention.
  • FIG. 3 A is a schematic diagram of the configuration of an example transform- based codec according to an embodiment of the invention.
  • FIG. 3B is a block diagram of an example decoder for intra-predicted macroblocks, according to an embodiment of the invention.
  • FIG. 3C is a block diagram of an example decoder for inter-predicted macroblocks, according to an embodiment of the invention.
  • FIG. 3D is a schematic diagram of the configuration of an example transform based codec employing feature -based prediction, according to an embodiment of the invention.
  • FIG. 4 is a block diagram of an example decoder within a feature -based prediction framework, according to an embodiment of the invention.
  • FIG. 5 is a block diagram illustrating the state isolation process of feature instances according to an embodiment of the present information.
  • FIG. 6 A is a block diagram illustrating example elements of a codec employing parametric modeling, according to an embodiment of the invention.
  • FIG. 6B is a block diagram illustrating example elements of parametric model- based adaptive encoder, according to an embodiment of the invention.
  • FIG. 6C is a block diagram illustrating the motion compensated prediction of features via interpolation of feature model parameters, according to an embodiment of the invention.
  • FIG. 7 A is a block diagram illustrating an overview of example cache architecture according to an embodiment of the invention.
  • FIG. 7B is a block diagram illustrating the processing involved in utilizing the local (short) cache data, according to an embodiment of the invention.
  • FIG. 7C is a block diagram illustrating the processing involved in utilizing the distant cache data, according to an embodiment of the invention.
  • FIG. 8 A is a schematic diagram of a computer network environment in which embodiments are deployed.
  • FIG. 8B is a block diagram of the computer nodes in the network of FIG. 8 A.
  • FIG. 8C is a screenshot of a feature-based compression tool in accordance with example implementations.
  • FIG. 8D is a screenshot showing features designated with numbers for both face and non-facial features, according to an embodiment of the invention.
  • FIG. 8E is a screenshot showing a face designated by the face tracker of FIG. 8D, according to an embodiment of the invention.
  • H.264 (sometimes used together with “compression,” “codecs,” “encodings,” or “encoders”) will refer to H.264, and “macroblocks” will be referred to without loss of generality as the basic H.264 coding unit.
  • Example elements of the invention may include video compression and decompression processes that can optimally represent digital video data when stored or transmitted.
  • the processes may include or interface with a video compression/encoding algorithm(s) to exploit redundancies and irrelevancies in the video data, whether spatial, temporal, or spectral. This exploitation may be done through the use and retention of feature-based models/parameters.
  • feature and “object” are used interchangeably.
  • Objects can be defined, without loss of generality, as "large features.” Both features and objects can be used to model the data.
  • FIG. 1A depicts a feature whose instances 10-1, 10-2,...,10-n have been detected in one or more frames of the video 20-1, 20-2,...,20-n.
  • a feature can be detected using several criteria based on both structural information derived from the pels and complexity criteria indicating that conventional compression utilizes a disproportionate amount of bandwidth to encode the feature region.
  • Each feature instance can be further identified spatially in its frame 20-1, 20-2, ...,20-n by a corresponding spatial extent or perimeter, shown in FIG. 1A as "regions" 30-1, 30-2, ..., 30-n.
  • These feature regions 30-1, 30-2, ..., 30-n can be extracted, for instance, as simple rectangular regions of pel data.
  • the feature regions are of size 16x16, the same size as H.264 macrob locks.
  • scale invariant feature transform (SIFT) [Lowe, David, 2004, “Distinctive image features from scale-invariant keypoints," Int. J. of Computer Vision, 60(2):91- 110] uses a convolution of a difference-of-Gaussian function with the image to detect blob-like features.
  • the speeded-up robust features (SURF) algorithm [Bay, Herbert et al, 2008, "SURF: Speeded up robust features,” Computer Vision and Image Understanding, 110(3):346-359] uses the determinant of the Hessian operator, also to detect blob-like features. In one embodiment of the present invention, the SURF algorithm is used to detect features.
  • features can be detected based on encoding complexity (bandwidth) encountered by a conventional encoder.
  • Encoding complexity for example, can be determined through analysis of the bandwidth (number of bits) required by conventional compression (e.g., H.264) to encode the regions in which features appear.
  • H.264 conventional compression
  • different detection algorithms operate differently, but each are applied to the entire video sequence of frames over the entire video data in embodiments.
  • a first encoding pass with an H.264 encoder is made and creates a "bandwidth map.” This in turn defines or otherwise determines where in each frame H.264 encoding costs are the highest.
  • each tile can be analyzed as a potential feature, based on the relative bandwidth required by H.264 to encode the tile. For example, the bandwidth required to encode a tile via H.264 may be compared to a fixed threshold, and the tile can be declared a "feature" if the bandwidth exceeds the threshold.
  • the threshold may be a preset value. The preset value may be stored in a database for easy access during feature detection. The threshold may be a value set as the average bandwidth amount allocated for previously encoded features .
  • the threshold may be a value set as the median bandwidth amount allocated for previously encoded features.
  • video frames can be partitioned into overlapping tiles. The overlapping sampling may be offset so that the centers of the overlapping tiles occur at the intersection of every four underlying tiles' corners. This over-complete partitioning is meant to increase the likelihood that an initial sampling position will yield a detected feature. Other, possibly more complex, topological partitioning methods are also possible.
  • Small spatial regions detected as features can be analyzed to determine if they can be combined based on some coherency criteria into larger spatial regions.
  • Spatial regions can vary in size from small groups of pels to larger areas that may correspond to actual objects or parts of objects. However, it is important to note that the detected features need not correspond to unique and separable entities such as objects and sub- objects.
  • a single feature may contain elements of two or more objects or no object elements at all.
  • the critical characteristic of a feature is that the set of pels comprising the feature can be efficiently compressed, relative to conventional methods, by feature model-based compression techniques.
  • Coherency criteria for combining small regions into larger regions may include: similarity of motion, similarity of appearance after motion compensation, and similarity of encoding complexity. Coherent motion may be discovered through higher-order motion models.
  • the translational motion for each individual small region can be integrated into an affine motion model that is able to approximate the motion model for each of the small regions. If the motion for a set of small regions can be integrated into aggregate models on a consistent basis, this implies a dependency among the regions that may indicate a coherency among the small regions that could be exploited through an aggregate feature model.
  • feature association is the basis for feature tracking (determining the location of a particular feature over time), described below.
  • feature association process must first define a feature model that can be used to discriminate similar feature instances from dissimilar ones.
  • the feature pels themselves can be used to model a feature.
  • Feature pel regions which are two-dimensional, can be vectorized and similar features can be identified by minimizing mean-squared error (MSE) or maximizing inner products between different feature pel vectors.
  • MSE mean-squared error
  • feature pel vectors are sensitive to small changes in the feature, such as translation, rotation, scaling, and changing illumination of the feature.
  • the invention accounts for such feature changes in the simplest way, by applying standard motion estimation and compensation algorithms found in conventional codecs (e.g., H.264), which account for translational motion of features.
  • standard motion estimation and compensation algorithms found in conventional codecs (e.g., H.264), which account for translational motion of features.
  • more complex techniques can be used to account for rotations, scalings, and illumination changes of features from frame to frame.
  • feature models are compact representations of the features themselves (“compact” meaning “of lower dimension than the original feature pels vectors”) that are invariant (remain unchanged when transformations of a certain type are applied) to small rotations, translations, scalings, and possibly illumination changes of the feature - meaning that if the feature changes slightly from frame to frame, the feature model will remain relatively constant.
  • a compact feature model of this type is often termed a "descriptor.”
  • the SURF feature descriptor has length 64 (compared to the length-256 feature pel vectors) and is based on sums of Haar wavelet transform responses.
  • a color histogram with 5 bins is constructed from a colormap of the feature pels, and this 5-component histogram acts as the feature descriptor.
  • feature regions are transformed via 2-D DCT. The 2-D DCT coefficients are then summed over the upper triangular and lower triangular portions of the coefficient matrix. These sums then comprise an edge feature space and act as the feature descriptor.
  • each instance of a feature that appears in multiple frames is a sample of the appearance of that feature, and multiple feature instances that are associated across frames are considered to "belong" to the same feature.
  • multiple feature instances belonging to the same feature may either be aggregated to form a feature track or gathered into an ensemble matrix 40 (FIG. 1A).
  • a feature track is defined as the (x,y) location of a feature as a function of frames in the video.
  • One embodiment associates newly detected feature instances with previously tracked features (or, in the case of the first frame of the video, with previously detected features) as the basis for determining which features instances in the current frame are extensions of which previously-established feature tracks.
  • the identification of a feature's instance in the current frame with a previously established feature track (or, in the case of the first video frame, with a previously detected feature) constitutes the tracking of the feature.
  • FIG. IB demonstrates the use of a feature tracker 70 to track features 60-
  • a feature detector 80 (for example, SIFT or SURF) is used to identify features in the current frame. Detected feature instances in the current frame 90 are matched to previously detected (or tracked) features 50.
  • the set of candidate feature detections in the current frame can be sorted using an auto-correlation analysis (ACA) metric that measures feature strength based on an autocorrelation matrix of the feature, using derivative-of-Gaussian filters to compute the image gradients in the autocorrelation matrix, as found in the Harris- Stephens corner detection algorithm [Harris, Chris and Mike Stephens, 1988, "A combined corner and edge detector," in Proc. of the 4th Alvey Vision Conference, pp. 147-151].
  • ACA auto-correlation analysis
  • Feature instances with high ACA values are given priority as candidates for track extension.
  • feature instances lower in the ACA-sorted list are pruned from the set of candidate features if they are within a certain distance (e.g., one pel) of a feature instance higher in the list.
  • feature descriptors e.g., the SURF descriptor
  • the feature pel vectors themselves may serve as the feature models.
  • previously-tracked features depicted as regions 60-1, 60-2, ..., 60-n in FIG. IB, are tested one at a time for track extensions from among the newly detected features in the current frame 90.
  • the most recent feature instance for each feature track serves as a focal point (or "target feature") in the search for a track extension in the current frame. All candidate feature detections in the current frame within a certain distance (e.g., 16 pels) of the location of the target feature are tested, and the candidate having minimum MSE with the target feature is chosen as the extension of that feature track.
  • a candidate feature is disqualified from being a track extension if its MSE with the target feature is larger than some threshold.
  • a limited search for a matching region in the current frame is conducted using either the motion compensated prediction (MCP) algorithm within H.264 or a generic motion estimation and compensation (MEC) algorithm.
  • MCP motion compensated prediction
  • MEC generic motion estimation and compensation
  • Both MCP and MEC conduct a gradient descent search for a matching region in the current frame that minimizes MSE (and satisfies the MSE threshold) with respect to the target feature in the previous frame. If no matches can be found for the target feature in the current frame, either from the candidate feature detection or from the MCP/MEC search process, the corresponding feature track is declared "dead" or "terminated.”
  • all but one of the feature tracks are pruned, or dropped from further consideration.
  • the pruning process keeps the feature track that has the longest history and has the largest total ACA, summed over all feature instances.
  • FPA feature point analysis
  • macroblocks in the video frame are thought of as features, registration of the features/macroblocks is done through the MCP engine found in H.264, and feature/macroblocks are associated using the inter-frame prediction metrics (such as sum of absolute transform differences [SATD]) of H.264; this combination is termed the macroblock cache (MBC) tracker.
  • the MBC tracker is differentiated from standard inter-frame prediction because certain parameters are different (for example, search boundaries are disabled, so that the MBC tracker conducts a wider search for matches) and because certain aspects of the matching process are different.
  • SURF detections are related to nearby macroblocks, and the macroblocks are associated and tracked using the MCP and inter- frame prediction engines of H.264; this combination is termed the SURF tracker.
  • multiple feature instances may be gathered into an ensemble matrix for further modeling.
  • the feature instances depicted as regions 30-1, 30-2, ..., 30-n, have been associated and identified as representing the same feature.
  • the pel data from the regions can then be vectorized and placed into an ensemble matrix 40, with the whole of the ensemble representing the feature.
  • This feature appearance model is of the same dimension as the feature and is distinct from the feature descriptor model described above.
  • the ensemble of regions can be spatially normalized (brought into conformity with a standard by removing sources of variation) toward one key region in the ensemble.
  • the region closest to the geometric centroid of the ensemble is selected as the key region.
  • the earliest feature instance in the ensemble is selected as the key region.
  • the appearance ensemble is processed to yield an appearance model
  • the deformation ensemble is processed to yield a deformation model.
  • the appearance and deformation models in combination become the feature model for the feature.
  • the feature model can be used to represent the feature with a compact set of parameters.
  • the method of model formation is singular value decomposition (SVD) of the ensemble matrix followed by a rank reduction in which only a subset of singular vectors and their corresponding singular values are retained.
  • the criterion for rank reduction is to retain just enough principal singular vectors (and corresponding singular values) that the reduced-rank reconstruction of the ensemble matrix approximates the full ensemble matrix to within an error threshold based on the 2-norm of the ensemble matrix.
  • the method of model formation is orthogonal matching pursuit (OMP) [Pati, Y.C. et al, 1993, "Orthogonal matching pursuit: Recursive function approximation with applications to wavelet decomposition," in Proc. of the 27th Asilomar Conference, pp. 40-44], wherein the ensemble is considered a pattern dictionary that is repeatedly searched to maximize reconstruction precision. Again, just enough ensemble vectors (and corresponding OMP weights) are retained such that the OMP reconstruction satisfies an error threshold based on the 2-norm of the ensemble matrix. Once formed, the appearance and deformation models of the feature can be used in feature-based compression, as will be described below.
  • OMP orthogonal matching pursuit
  • the feature ensemble can be refined by comparing ensemble members to each other.
  • the ensemble is refined by exhaustively comparing each sampled region (ensemble vector) to every other sampled region.
  • This comparison is comprised of two tile registrations.
  • One registration is a comparison of a first region to a second region.
  • the second registration is a comparison of the second region to the first region.
  • Each registration is performed at the position of the regions in their respective images.
  • the resulting registration offsets, along with the corresponding positional offsets, are retained and referred to as correlations.
  • the correlations are analyzed to determine if the multiple registrations indicate that a sampled region's position should be refined. If the refined position in the source frame yields a lower error match for one or more other regions, then that region position is adjusted to the refined position.
  • the refined position of the region in the source frame is determined through a linear interpolation of the positions of other region correspondences that temporally span the region in the source frame.
  • Standard inter-frame prediction uses block-based motion estimation and compensation to find predictions for each coding unit
  • feature modeling provides a way of navigating the prediction search space to produce improved predictions without prohibitive computations.
  • feature-based and “model-based” are used interchangeably, as features are a specific type of model.
  • feature tracks are used to relate features to macrob locks. The general steps for this are depicted in FIG. 1C.
  • a given feature track indicates the location of a feature across frames, and there is an associated motion of that feature across frames. Using the location of the feature in the two most recent frames prior to the current frame, one can project the position of the feature in the current frame. This projected feature position then has an associated nearest
  • macrob lock defined as the macrob lock having greatest overlap with the projected feature position.
  • This macroblock (now the target macroblock that is being encoded) has been associated to a specific feature track whose projected position in the current frame is nearby the macroblock (100 in FIG. 1C).
  • the next step is to calculate an offset 110 between the target macroblock and the projected feature position in the current frame.
  • This offset can then be used to generate predictions for the target macroblock, using earlier feature instances in the associated feature's track. These earlier feature instances occupy either a local cache 120, comprised of recent reference frames where the feature appeared, or a distant cache 140, comprised of "older" reference frames 150 where the feature appeared.
  • Predictions for the target macroblock can be generated by finding the regions in the reference frames with the same offsets (130, 160) from earlier feature instances as the offset between the target macroblock and the projected feature position in the current frame.
  • feature -based prediction is implemented as follows: (1) detect the features for each frame; (2) model the detected features; (3) associate features in different frames to create feature tracks; (4) use feature tracks to predict feature locations in the "current" frame being encoded; (5) associate macroblocks in the current frame that are nearby the predicted feature locations; (6) generate predictions for the macroblocks in Step 5 based on past locations along the feature tracks of their associated features.
  • features are detected using the SURF algorithm and they are associated and tracked using the FPA algorithm, as detailed in the previous section. Once features have been detected, associated, and tracked, the feature tracks can be used to associate each feature track with a nearest macroblock, as detailed above. It is possible for a single macroblock to be associated with multiple features, so one embodiment selects the feature having maximum overlap with the macroblock as the associated feature for that macroblock.
  • a primary or key prediction for the target macroblock can be generated.
  • Data (pels) for the key prediction comes from the most recent frame (prior to the current frame) where the feature appears, henceforth referred to as the key frame.
  • the key prediction is generated after selecting a motion model and a pel sampling scheme.
  • the motion model can be either "0th order,” which assumes that the feature is stationary between the key frame and the current frame, or "1st order,” which assumes that feature motion is linear between the 2nd-most recent reference frame, the key frame, and the current frame.
  • the motion of the feature is applied (in the backwards temporal direction) to the associated macroblock in the current frame to obtain the prediction for the macroblock in the key frame.
  • the pel sampling scheme can be either "direct,” in which motion vectors are rounded to the nearest integer and pels for the key prediction are taken directly from the key frame, or "indirect,” in which the interpolation scheme from conventional compression such as H.264 is used to derive a motion-compensated key prediction.
  • the present invention can have four different types of key prediction, depending on the motion model (0 th or 1 st order) and the sampling scheme (direct or indirect).
  • Key prediction can be refined by modeling local deformations through the process of subtiling.
  • subtiling In the subtiling process, different motion vectors are calculated for different local portions of the macroblock.
  • subtiling can be done by dividing the 16x16 macroblock into four 8x8 quadrants and calculating predictions for each separately.
  • subtiling can be carried out in the Y/U/V color space domain by calculating predictions for the Y, U, and V color channels separately.
  • the offset from the target macroblock to the (projected) position of the associated feature in the current frame represents a motion vector that can be used to find secondary predictions from the feature's position in past reference frames. In this way, a large number of secondary predictions can be generated (one for each frame where the feature has appeared previously) for a given target macroblock that has an associated feature.
  • the number of secondary predictions can be limited by restricting the search to some reasonable number of past reference frames (for example, 25).
  • the overall reconstruction of the target macroblock can be computed based on these predictions.
  • the reconstruction is based on the key prediction only, henceforth referred to as key-only (KO) reconstruction.
  • the reconstruction is based on a composite prediction that sums the key prediction and a weighted version of one of the secondary predictions.
  • This algorithm henceforth referred to as PCA-Lite (PCA-L), involves the following steps:
  • the steps in the PCA-Lite algorithm approximate the operations in the well-known orthogonal matching pursuit algorithm [Pati, 1993], with the composite prediction meant to have non-redundant contributions from the primary and secondary predictions.
  • the PCA-Lite algorithm described above is modified so that the key vector in Steps 3-5 above is replaced by the mean of the key and the secondary vector. This modified algorithm is henceforth referred to as PCA- Lite-Mean.
  • PCA-Lite provides a different type of composite prediction than the bi-prediction algorithms found in some standard codecs (and described in the "Background" section above).
  • Standard bi-prediction algorithms employ a blending of multiple predictions based on temporal distance of the reference frames for the individual predictions to the current frame.
  • PCA-Lite blends multiple predictions into a composite prediction based on the contents of the individual predictions.
  • composite predictions can be formed from any set of multiple predictions for a given target macroblock.
  • Feature -based modeling provides a naturally-associated set of multiple predictions for a given target
  • the current invention provides the ability to model the data at multiple fidelities for the purpose of model-based compression.
  • FIG. 2A displays four levels of modeling. These four levels are summarized in the following table and discussed in more detail below.
  • the bottom level 200 in FIG. 2A is termed the "Macroblock” (MB) level and represents conventional compression partitioning frames into non-overlapping macroblocks, tiles of size 16x16, or a limited set of subtiles.
  • Conventional compression e.g., H.264
  • BBMEC block-based motion estimation and compensation
  • the predictions 212 are combined with residual encodings of the macrob locks (or subtiles) to synthesize 210 a reconstruction of the original data.
  • the second level 202 in FIG. 2A is termed the "Macroblocks as Features"
  • MBC MBC level
  • macroblocks or subtiles of macroblocks
  • the first application of BBMEC is identical to that of the MB level, finding a conventional prediction for the target macrob lock from the most recent reference frame in 216.
  • the second application of BBMEC finds a conventional prediction for the first prediction by searching in the second-most-recent frame in 216.
  • Repeated application of BBMEC through progressively older frames in 216 creates a "track" for the target macrob lock, even though the latter has not been identified as a feature per se.
  • the MBC track produces a model 214 that generates a prediction 212 that is combined with residual encodings of the macroblocks (or subtiles) to synthesize 210 a reconstruction of the original data at the decoder.
  • the third level 204 in FIG. 2A is termed the "Features" level and represents feature-based compression as described above. To review, features are detected and tracked independent of the macroblock grid, but features are associated with
  • overlapping macroblocks and feature tracks are used to navigate previously-decoded reference frames 216 to find better matches for those overlapping macroblocks. If multiple features overlap a given target macroblock, the feature with greatest overlap is selected to model that target macroblock at 214.
  • the codec could encode and decode the features directly, without relating the features to macroblocks, and process the "non-feature" background separately using, for example, MB-level conventional compression.
  • the feature-based model 214 generates a prediction 212 that is combined with residual encodings of the associated macrob locks (or subtiles) to synthesize 210 a reconstruction of the original data at the decoder.
  • the top level 206 in FIG. 2A is termed the "Objects" level and represents object-based compression.
  • Objects are essentially large features that may encompass multiple macrob locks and may represent something that has physical meaning (e.g., a face, a ball, or a cellphone) or complex phenomena 208.
  • Object modeling is often parametric, where it is anticipated that an object will be of a certain type (e.g., a face), so that specialized basis functions can be used for the modeling 214.
  • a single motion vector 212 can be calculated for all of the macroblocks associated with the object 216, which can result in savings both in terms of computations and encoding size.
  • the object-based model 214 generates a prediction 212 that is combined with residual encodings of the associated macroblocks (or subtiles) to synthesize 210 a reconstruction of the original data at the decoder.
  • objects may also be identified by correlating and aggregating nearby feature models 214.
  • FIG. 2B is a block diagram illustrating this type of nonparametric or empirical object detection via feature model aggregation. A particular type of object 220 is detected by identifying which features have
  • object bias 222
  • the set of features in 222 display a rigidity of the model states 224, a tendency over time for the features and their states to be correlated. If the individual feature models are determined to be correlated (in which case an object detection is determined 226), then a composite appearance model with accompanying parameters 228 and a composite deformation model with accompanying parameters 230 can be formed. The formation of composite appearance and deformation models evokes a natural parameter reduction 232 from the collective individual appearance and deformation models.
  • FIG. 2C illustrates a third embodiment of the "Objects" level 206 in FIG. 2A, employing both parametric and nonparametric object-based modeling.
  • a parametrically modeled object is detected 240.
  • the detected object 240 may be processed to determine if there are any overlapping features 250.
  • the set of overlapping features may then be tested 260 to determine whether they can be aggregated as above. If aggregation of the overlapping features fails, then the process reverts to testing the macroblocks overlapping the detected object 240, to determine whether they can be effectively aggregated 270 to share a common motion vector, as noted above.
  • a multiple-fidelity processing architecture may use any combination of levels 200, 202, 204, 206 to achieve the most advantageous processing.
  • all levels in FIG. 2A are examined in a "competition" to determine which levels produce the best (smallest) encodings for each macroblock to be encoded. More details on how this "competition" is conducted follow below.
  • the levels in FIG. 2 A could be examined sequentially, from bottom (simplest) to top (most complex). If a lower-level solution is deemed satisfactory, higher-level solutions do not have to be examined. Metrics for determining whether a given solution can be deemed "good enough" are described in more detail below.
  • the encoding process may convert video data into a compressed, or encoded, format.
  • the decompression process, or decoding process may convert compressed video back into an uncompressed, or raw, format.
  • the video compression and decompression processes may be implemented as an encoder/decoder pair commonly referred to as a codec.
  • FIG. 3A is a block diagram of a standard encoder 312.
  • the encoder in FIG. 3A may be implemented in a software or hardware environment, or combination thereof. Components of the example encoder may be implemented as executable code stored on a storage medium, such as one of those shown in FIGs. 8 A and 8B, and configured for execution by one or more of processors 820.
  • the encoder 312 may include any combination of components, including, but not limited to, an intra-prediction module 314, an inter-prediction module 316, a transform module 324, a quantization module 326, an entropy encoding module 328 and a loop filter 334.
  • the inter prediction module 316 may include a motion compensation module 318, frame storage module 320, and motion estimation module 322.
  • the encoder 312 may further include an inverse quantization module 330, and an inverse transform module 332.
  • the function of each of the components of the encoder 312 shown in FIG. 3 A is well known to one of ordinary skill in
  • the entropy coding algorithm 328 in FIG. 3 A may be based on a probability distribution that measures the likelihood of different values of quantized transform coefficients.
  • the encoding size of the current coding unit (e.g., macroblock) depends on the current encoding state (values of different quantities to be encoded) and the relative conformance of the state to the probability distribution. Any changes to this encoding state, as detailed below, may impact encoding sizes of coding units in subsequent frames.
  • an exhaustive search may be conducted of all the possible paths on which the video can be encoded (i.e., all possible encoding states), but this is computationally prohibitive.
  • the encoder 312 is configured to focus on the current (target) macroblock, so that optimization is applied locally, rather than considering a larger scope, (e.g., over a slice, a frame, or a set of frames).
  • FIGs. 3B and 3C are block diagrams of a standard decoder 340 providing decoding of intra-predicted data 336 and decoding of inter-predicted data 338, respectively.
  • the decoder 340 may be implemented in a software or hardware environment, or combination thereof.
  • the encoder 312 typically receives the video input 310 from an internal or external source, encodes the data, and stores the encoded data in the decoder cache/buffer 348.
  • the decoder 340 retrieves the encoded data from the cache/buffer 348 for decoding and transmission.
  • the decoder may obtain access to the decoded data from any available means, such as a system bus or network interface.
  • the decoder 340 can be configured to decode the video data to decompress the predicted frames and key frames (generally at 210 in FIG. 2A).
  • the cache/buffer 348 can receive the data related to the compressed video sequence/bitstream and make information available to the entropy decoder 346.
  • the entropy decoder 346 processes the bitstream to generate estimates of quantized transform coefficients for the intra-prediction in FIG. 3A or the residual signal in FIG. 3B.
  • the inverse quantizer 344 performs a rescaling operation to produce estimated transform coefficients, and the inverse transform 342 is then applied to the estimated transform coefficients to create a synthesis of the intra-prediction of the original video data pels in FIG. 3A or of the residual signal in FIG.
  • the synthesized residual signal is added back to the inter-prediction of the target macroblock to generate the full reconstruction of the target macroblock.
  • the inter-prediction module 350 replicates at the decoder the inter-prediction generated by the encoder, making use of motion estimation 356 and motion compensation 354 applied to reference frames contained in the framestore 352.
  • the decoder's inter-prediction module 350 mirrors the encoder's inter-prediction module 316 in FIG. 3 A, with its components of motion estimation 322, motion compensation 318, and framestore 320.
  • Hybrid codec implementing model-based prediction
  • FIG. 3D is a diagram of an example encoder according to an embodiment of the invention that implements model-based prediction.
  • the codec 360 can be configured to encode a current (target) frame.
  • each macroblock in the frame can be encoded, such that, at 366, a standard H.264 encoding process is used to define a base (first) encoding that yields an H.264 encoding solution.
  • the encoder 366 is an H.264 encoder capable of encoding a Group of Pictures (set of reference frames). Further, the H.264 encoder preferably is
  • inter-frame prediction able to search multiple reference frames for good matches for the macroblock being encoded.
  • the error between the original macroblock data and the prediction is transformed, quantized, and entropy-encoded.
  • the encoder 360 utilizes the CAB AC entropy encoding algorithm at
  • the context modeling may be applied to a binarized sequence of the syntactical elements of the video data such as block types, motion vectors, and quantized coefficients, with the binarization process using predefined mechanisms. Each element is then coded using either adaptive or fixed probability models. Context values can be used for appropriate adaptations of the probability models.
  • the H.264 macroblock encoding is analyzed.
  • efficiency of the H.264 encoding can be judged by comparing the H.264 encoding size (in bits) to a threshold, which can be derived from percentile statistics from previously encoded videos or from earlier in the same video. In another embodiment, efficiency of the
  • H.264 encoding can be judged by determining whether an H.264 encoder has declared the target macroblock a "skip" macroblock, in which the data in and around the target macroblock is uniform enough that the target macroblock essentially requires no additional encoding.
  • the encoder enters Competition Mode 380.
  • Competition Mode 380 In this mode, several different predictions are generated for the target macroblock, based on multiple models 378.
  • the models 378 are created from the identification of features 376 detected and tracked in prior frames 374. Note that as each new frame 362 is processed (encoded and then decoded and placed into framestore), the feature models need to be updated to account for new feature detections and associated feature track extensions in the new frame 362.
  • the model-based solutions 382 are ranked based on their encoding sizes 384, along with the H.264 solution acquired previously. Because of its flexibility to encode a given macroblock using either a base encoding (the H.264 solution) or a model-based encoding, the present invention is termed a hybrid codec.
  • an H.264 encoding is generated for the target macroblock to compare its compression efficiency (ability to encode data with a small number of bits) relative to other modes. Then for each encoding algorithm used in Competition Mode, the following steps are executed: (1) generate a prediction based on the codec mode/algorithm used; (2) subtract the prediction from the target macroblock to generate a residual signal; (3) transform the residual (target minus prediction) using an approximation of a 2-D block-based DCT; (4) encode the transform coefficients using an entropy encoder.
  • the baseline H.264 (inter-frame) prediction can be thought of as based on a relatively simple, limited model (H.264 is one of the algorithms used in Competition Mode).
  • H.264 is one of the algorithms used in Competition Mode.
  • the predictions of the encoder 360 can be based on more complex models, which are either feature-based or object-based, and the corresponding tracking of those models. If a macroblock exhibiting data complexity is detected, the encoder 360 operates under the assumption that feature-based compression can do a better job than conventional compression.
  • the solution search space for a given target macroblock is comprised of all of the invention's feature-based predictions represented above, plus the H.264 solution (the "best" inter-frame prediction from H.264).
  • Competition Mode includes all possible combinations of processing choices noted above (tracker type, motion model and sampling scheme for key prediction, subtiling scheme, and reconstruction algorithms).
  • Competition Mode are configurable and can be limited to a reasonable subset of possible processing combinations to save computations.
  • Potential solutions for the competition are evaluated one at a time by following the four steps noted previously: (1) generate the prediction; (2) subtract the prediction from the target macroblock to generate a residual signal; (3) transform the residual; (4) encode the transform coefficients using an entropy encoder.
  • the output of the last step, 382 is a number of bits associated with a given solution 384.
  • the encoder is rolled back to its state prior to that evaluation, so that the next solution can be evaluated.
  • a "winner" for the competition is chosen 370 by selecting the one with smallest encoding size. The winning solution is then sent to the encoder once more 372 as the final encoding for the target macroblock.
  • this winning solution is a locally-optimum solution, as it is optimum for the target macroblock only.
  • the selection of the optimal solution is hedged against larger scale encoding tradeoffs that include, but are not limited to, context intra-frame prediction feedback and residual error effects in future frames.
  • Information pertaining to the winning solution is saved into the encoding stream 386 and transmitted/stored for future decoding.
  • This information may include, but is not limited to, the processing choices noted above for feature-based prediction (e.g., tracker type, key calculation, subtiling scheme, reconstruction algorithm, etc.).
  • the encoder 360 may determine that the target macroblock is not efficiently coded by H.264, but there is also no detected feature that overlaps with that macroblock. In this case, the encoder uses H.264 anyway to encode the macroblock as a last resort.
  • the tracks from the feature tracker can be extended to generate a pseudo-feature that can overlap the macroblock and thus produce a feature-based prediction.
  • movement among the four levels in Fig. 2A is governed by Competition Mode.
  • FIG. 4 is a diagram of an example decoder according to an embodiment of the invention implementing model-based prediction within the Assignee's EuclidVision codec.
  • the decoder 400 decodes the encoded video bitstream to synthesize an approximation of the input video frame that generated the frame encoding 402.
  • the frame encoding 402 includes a set of parameters used by the decoder 400 to reconstruct its corresponding video frame 418.
  • the decoder 400 traverses each frame with the same slice ordering used by the encoder, and the decoder traverses each slice with the same macroblock ordering used by the encoder. For each macroblock 404, the decoder follows the same process as the encoder, determining 406 whether to decode the macroblock conventionally 408 or whether to decode the macroblock utilizing feature models and parameters at 416. If a macroblock was encoded via the invention's model-based prediction, the decoder 400 extracts whatever feature information (feature tracks, feature reference frames [GOP], feature motion vectors) needed to reproduce the prediction for that solution 418. The decoder updates feature models (410, 412, 414) during the decoding so they are synchronized with the encoder feature state for the particular frame/slice/macroblock that is being processed.
  • feature models feature models
  • FIG. 5 is a block diagram illustrating the state isolation process 500 of feature instances according to an embodiment of the present invention.
  • This state isolation information can be associated with a target macroblock and include parameters associated with relevant feature instances 502 that can be of assistance in the encoding of that target macroblock.
  • the state isolation information can be also used to interpolate predicted features in future video frames.
  • Each respective feature instance has an associated GOP 504.
  • Each GOP includes respective state information regarding, for example, respective boundary information.
  • the respective state isolation information of a feature instance may further include state information about any relevant associated objects, their respective slice parameters 506, and their respective entropy state 508.
  • the state information provides instructions regarding the boundaries of GOP/slice/entropy parameters of feature instances and their corresponding extensions into new states and state contexts.
  • the state information 506, 508 can be used to predict and interpolate the state of a predicted feature in future frames.
  • the extended prediction context for the encoder 312 in FIG. 3 A and decoder 340 in FIGs. 3B and 3C may include, but is not limited to: (1) one or more macroblocks, (2) one or more neighboring macroblocks, (3) slice information, (4) reference frames
  • feature models may be used explicitly in the codec framework.
  • Specific regions in the target frame can be represented by certain types of models (for example, face models), and the representation is dependent on the parameters in the models.
  • This type of explicit modeling is henceforth referred to as parametric modeling
  • the codec implementation described in the above section uses nonparametric or empirical modeling. Because parametric modeling expects certain types of features or objects (e.g., faces), the modeling usually consists of a set of basis vectors that span the space of all possible features/objects of that type, and the model parameters are the projections of the target region onto the basis functions.
  • FIG. 6 A is a block diagram illustrating example elements of a codec 600 according to an alternative embodiment of the invention implementing parametric modeling.
  • the codec 600 in FIG. 6A may include modules to perform adaptive motion compensated prediction 610, adaptive motion vector prediction 612, adaptive transform processing 614, and/or adaptive entropy encoding 616.
  • the adaptive motion compensation module 610 can be configured to select reference frames 618 based on frames having instances of features. If models of the features provide improved compression efficiency, then the frames from which those models were derived can be selected as reference frames, and an associated Group of Pictures may be generated. Interpolation of the motion vector offsets 626 may be performed based on the parameters from the detected features.
  • new data pels for a predicted feature instance may be constructed within the range of a discrete set of known data points based on previously detected features.
  • Subtile partitioning processing 612 decisions in the conventional encoder are supplemented by the constraints of deformation variation models 620.
  • Transform processing 614 can be adapted to utilize appearance variation modeling 622 to constrain appearance variation parameters.
  • Entropy encoding processing 616 can be supplemented by parametric range/scale analysis 624 and adaptive quantization 628 in the inventive codec 600.
  • the resulting macroblock supplementary data 630 is outputted by codec 600.
  • parametric modeling can be used to improve the predictions provided by the original hybrid codec described above.
  • elements of a parametric model are applied to an existing target macroblock prediction (such as, for example, the output of Competition Mode above) to determine whether the prediction can be improved.
  • FIG. 6B illustrates one application of a parametric model-based adaptive encoder 634.
  • the adaptive encoder 634-1 can be configured to supplement the encoding that would be performed in a conventional codec, such as H.264, or in a hybrid codec such as that described above.
  • the pel residual 636 resulting from a conventional motion compensation prediction process is analyzed 638 to determine if the deformation and appearance variation of the residual can be modeled 642 more efficiently by a parametric feature model.
  • one can determine the relative efficiency of the parametric model by measuring the reduction in the sum of absolute transform differences (SATD) 640 between the prediction residual 636 and the parametric model 638. If the parametric model is determined to be an efficient representation, the target region (macroblock) is projected onto the feature model (appearance and deformation bases), yielding features parameters that serve as the encoding of the residual signal.
  • SSATD sum of absolute transform differences
  • Additional rollback capabilities 634-2 are provided by this embodiment to test the applicability of the alternate residual modeling within the current GOP, slice, and entropy state.
  • reference frames 644, GOPs, and features (slices) 646 that are remote in the video frame sequence to the current frame being encoded in the series can be considered for references in prediction, whereas with conventional encoding this would not be practical.
  • the rollback may come from other video data, such as other video files, if feature models from those other video files provide improved compression.
  • the invariant components are certain parameters of the feature model (for example, coefficients that represent weightings of different basis functions).
  • the invariant components are typically the feature pels themselves.
  • the preservation of invariant model components can serve as a guiding principle (henceforth referred to as the "invariance principle") for how feature motion estimation and compensation is performed.
  • FIG. 6C is a block diagram illustrating motion compensated prediction of features via interpolation of feature model parameters, guided by the invariance principle, according to an embodiment of the invention.
  • the motion compensated prediction process 668 begins by adjusting model parameters from several feature instances toward an invariant instance of the parameters, a process known as normalization.
  • the collection 670 of feature instances (“matched macroblocks") can be used 672 to generate several types of interpolation functions (674, 676, 678, 680) for normalizing the instances toward the invariant instance.
  • the invariant instance of the model parameters 682 can be defined as the set of model parameter values in the key frame.
  • the invariant instance is then representative of most of (if not all) the predictions/patterns in the feature-based model.
  • the invariant instance is conceptually similar to the centroid of a vector space comprised by the vectorized form of the instances' appearance parameters.
  • the invariant instance 682 can then serve as the key pattern on which to extrapolate the target's position 684 using one of the interpolation functions (674, 676, 678, 680).
  • This interpolation/extrapolation process can be used to predict the frame position, appearance variation, and deformation variation of the feature in the target frame.
  • the combination of the invariant representation of the features with a compact parametric form of the feature instances represents a drastic reduction in the amount of memory required to cache the appearance and deformation of features contained in source reference frames as compared with conventional compression. In other words, the data in the frame that is relevant and useful for compression is captured concisely in the feature models.
  • the feature model parameters from two or more feature instances can be used to predict the state of the target region given the known temporal interval between reference frames where the feature instances occurred and the current (target) frame.
  • a state model an extrapolation of two or more feature parameters given temporal steps, can be used to predict the feature parameters for the target region, following the invariance principle.
  • the state model can be linear or higher-order (for example, an extended Kalman filter).
  • the feature model information can be stored or cached efficiently by organizing the model information prior to caching. This technique can be applied to both parametric and nonparametric model-based compression schemes.
  • the cache 348 (including the framestore 352) can be configured to include feature -based modeling prediction context information. Attempts to access uncached feature-based prediction context data can generate overhead that degrades the system's responsiveness and determinism. This overhead can be minimized by caching, ahead of time, the preprocessed feature-based encoding prediction context. Doing this provides a means by which much of the repetition of accessing data related to the feature-based prediction context can be avoided.
  • the encoder 312/decoder 340 can be configured using, for example, a cache that is adapted to increase the execution speed and efficiency of video processing.
  • the performance of the video processing may depend upon the ability to store, in the cache, feature-based encoding prediction data such that it is nearby in the cache to the associated encoded video data, even if that encoded video data is not spatially close to the frame(s) from which the feature-based encoding prediction data was originally derived.
  • Cache proximity is associated with the access latency, operational delay, and transmission times for the data.
  • the encoder 312/decoder 340 may include a configurator that stores the prediction data in the cache in such a way to ensure that, when a macroblock or frame is decoded, the feature- based prediction context information is easily accessible from the cache/buffer/ framestore .
  • Certain embodiments of the present invention can extend the cache by first defining two categories of feature correlation in the previously decoded frames, namely local and non-local previously decoded data for the cache.
  • the local cache can be a set of previously decoded frames that are accessible in batches, or groups of frames, but the particular frames that constitute those groups are determined by detected features.
  • the local cache is driven by features detected in the current frame.
  • the local cache is used to a greater extent when there are relatively few "strong" feature models (models having a long history) for the current frame/macroblock.
  • the local cache processing is based on batch motion compensated prediction, and groups of frames are stored in reference frame buffers.
  • FIG. 7A is a block diagram illustrating an overview of example cache architecture 710-1 according to an embodiment of the invention.
  • the cache access architecture 710-1 includes the decision processes 710 for local cache access 712 (716, 718, 720, 722, and 724) and distant cache access 714 (726, 728, 730, and 732). If the features are mostly local 712 (for example, there are few strong feature models for the current frame/macroblock), then local cache processing 718 is provided.
  • FIG. 7B is a block diagram illustrating the processing involved in utilizing the local (short) cache data 734.
  • the local cache can be a set of previously decoded frames that are accessible in batches, or groups of frames, but the particular frames that constitute those groups are determined by detected features.
  • the local cache 734 in FIG. 7B groups only "short history" features, those whose tracks only comprise a small number of frames.
  • the aggregate set of frames encompassed by the short history features determines a joint frameset 738 for those features.
  • Frames in the joint frameset 738 may be prioritized 740 based on the complexity of the feature tracks in the respective frames. In one embodiment, complexity may be determined by the encoding cost of the features from a base encoding process such as H.264. Referring to FIGs.
  • the local cache may be stored in the framestore 352 or in the cache buffer 348.
  • the locally cached frames are utilized at 720.
  • a GOP/batch 742 based on detected feature instances can then be formed at 722.
  • the GOP/batch based on detected feature instances can be tested at 724 as reference frames 744 for the motion
  • Motion compensated prediction done in this way can be said to be “biased” toward feature tracking information, because the reference frames for the motion estimation are the frames with previously-detected feature instances.
  • additional rollback capabilities are provided to test the applicability of the residual modeling within the GOP/batch, slice, and entropy state. In this way, reference frames that are remote in the video frame sequence to the current frame being encoded can be evaluated more efficiently.
  • certain embodiments of the invention are able to apply analysis to past frames to determine the frames that will have the highest probability of providing matches for the current frame.
  • the number of reference frames can be much greater than the typical one-to-sixteen reference frame maximum found in conventional compression.
  • the reference frames may number up to the limit of system memory, assuming that there are a sufficient number of useful matches in those frames.
  • the intermediate form of the data generated by the present invention can reduce the required amount of memory for storing the same number of reference frames.
  • the non-local cache is based on two different cache access methods, frame and retained.
  • the frame access of the nonlocal cache accesses frames directly to create feature models that are then utilized to encode the current frame.
  • the retained mode does not access the previously decoded data directly, but rather utilizes feature models that have been retained as data derived from those previously decoded frames (the feature model and the parameters of the instances of the feature model in those frames) and thereby can be used to synthesize that same data.
  • the models for the feature instances are accessed.
  • the reference frames are accessed, and at 732 the combination of optimal reference frames and models are marked for use. Criteria for optimality are based on intermediate feature information for the feature models in each reference frame, including feature strength and feature bandwidth.
  • the distant cache 714 can be any previously decoded data (or encoded data) that is preferably accessible in the decoder state.
  • the cache may include, for example, reference frames/GOPs, which are generally a number of frames that precede the current frame being encoded.
  • the decoder cache allows for other combinations of previously decoded frames to be available for decoding the current frame.
  • FIG. 7C is a block diagram illustrating the processing involved in utilizing the distant cache data.
  • the distant (non-local) cache 748 illustrates the longer range cache architecture.
  • the distant cache is initialized from the local cache 750 in response to a determination 752 that the detected features have an extensive history, both in terms of reoccurring instances and the repeated applicability of the correspondence models associated with those features.
  • the process determines which retention mode 754 is used.
  • the two modes of the non-local cache are the retained 760 and non-retained 756.
  • the non-retained 756 is a conventional motion compensated prediction process augmented with predictions based on feature models (similar to the usage of implicit modeling for the hybrid codec described above).
  • the non-retained mode 756 thus accesses 758 reference frames to obtain working predictions.
  • the retained mode is similar to the non-retained mode, but it uses predictions that come explicitly from the feature model itself 762, 766.
  • the retained model necessarily limits the prediction searches to that data for which the feature model is able to synthesize the feature that it models.
  • the feature model may contain the instance parameterizations for the feature's instances in prior frames, which would be equivalent to the pels contained in those prior frames.
  • the interpolation of the function describing those parameters is also used to provide predictions to the motion compensation prediction process to facilitate frame synthesis 764.
  • subsets can be selected using SVD, for example.
  • a subset of feature instances acts as a basis for the ensemble and can be cached and used to encode the corresponding feature whenever it occurs in subsequent frames of the video (or in other videos).
  • This subset of feature instances models the feature both compactly and accurately.
  • Example implementations of the present invention may be implemented in a software, firmware, or hardware environment.
  • FIG. 8A illustrates one such environment.
  • Client computer(s)/devices 810 and a cloud 812 (or server computer or cluster thereof) provide processing, storage, and input/output devices executing application programs and the like.
  • Client computer(s)/devices 810 can also be linked through communications network 816 to other computing devices, including other client devices/processes 810 and server computer(s) 812.
  • Communications network 816 can be part of a remote access network, a global network (e.g., the
  • Internet a worldwide collection of computers, Local area or Wide area networks, and gateways that currently use respective protocols (TCP/IP, Bluetooth, etc.) to
  • FIG. 8B is a diagram of the internal structure of a computer/computing node (e.g., client processor/device 810 or server computers 812) in the processing
  • Each computer 810, 812 contains a system bus 834, where a bus is a set of actual or virtual hardware lines used for data transfer among the components of a computer or processing system.
  • Bus 834 is essentially a shared conduit that connects different elements of a computer system (e.g., processor, disk storage, memory, input/output ports, etc.) that enables the transfer of information between the elements.
  • Attached to system bus 834 is an I/O device interface 818 for connecting various input and output devices (e.g., keyboard, mouse, displays, printers, speakers, etc.) to the computer 810, 812.
  • Network interface 822 allows the computer to connect to various other devices attached to a network (for example the network illustrated at 816 of FIG. 8A).
  • Memory 830 provides volatile storage for computer software instructions 824 and data 828 used to implement an embodiment of the present invention (e.g., codec, video encoder/decoder code).
  • Disk storage 832 provides nonvolatile storage for computer software instructions 824 (equivalently, "OS program” 826) and data 828 used to implement an embodiment of the present invention; it can also be used to store the video in compressed format for long-term storage.
  • Central processor unit 820 is also attached to system bus 834 and provides for the execution of computer instructions. Note that throughout the present text, "computer software instructions” and "OS program” are equivalent.
  • the processor routines 824 and data 828 are a computer program product (generally referenced 824), including a computer readable medium capable of being stored on a storage device 828, which provides at least a portion of the software instructions for the invention system.
  • the computer program product 824 can be installed by any suitable software installation procedure, as is well known in the art.
  • at least a portion of the software instructions may also be downloaded over a cable, communication, and/or wireless connection.
  • the invention programs are a computer program propagated signal product 814 (in Fig 8 A) embodied on a propagated signal on a propagation medium (e.g., a radio wave, an infrared wave, a laser wave, a sound wave, or an electrical wave propagated over a global network such as the Internet, or other network(s)).
  • a propagation medium e.g., a radio wave, an infrared wave, a laser wave, a sound wave, or an electrical wave propagated over a global network such as the Internet, or other network(s)
  • a propagation medium e.g., a radio wave, an infrared wave, a laser wave, a sound wave, or an electrical wave propagated over a global network such as the Internet, or other network(s)
  • Such carrier media or signals provide at least a portion of the software instructions for the present invention routines/program 824, 826.
  • the propagated signal is an analog carrier wave or digital signal carried on the propagated medium.
  • the propagated signal is transmitted over the propagation medium over a period of time, such as the instructions for a software application sent in packets over a network over a period of milliseconds, seconds, minutes, or longer.
  • the computer readable medium of computer program product 824 is a propagation medium that the computer system 810 may receive and read, such as by receiving the propagation medium and identifying a propagated signal embodied in the propagation medium, as described above for computer program propagated signal product.
  • FIG. 8C is a screenshot 840 of a feature -based display tool in accordance with example implementations.
  • Screenshot 840 shows a frame of video with features identified in boxes 842.
  • the video frame sequence context for the frame is identified at 844.
  • the features 842 were tracked through the frames 844, creating several feature sets that are represented in section 846 of the display.
  • Within a particular feature set 846 there is a plurality of feature members (feature instances) 848.
  • a data area displays the feature bandwidth 852, the number of bits required by conventional compression to code the feature.
  • An indication 850 of the feature detection process is also displayed in the data area.
  • the tool displays all features and feature tracks that were identified in the subject video.
  • FIG. 8E is a screenshot 860-2 showing a face 864 designated by the face tracker.
  • FIG. 8D is a screenshot 860-1 showing features designated with numbers 862 for both face and non-facial features. In this example, the numbers shown in FIG. 8D represent the length of tracking of the features through frames.
  • All pels/pixels within a region of interest may be encoded using the face model instead of strictly using an H.264 encoder process.
  • face model biasing is not needed, and H.264 is not used to select prior reference frames.
  • the face is generated based on the feature correspondence models, and then lower level processing is used to encode the residual.
  • the models of the present invention can be used as a way to control access to the encoded digital video. For example, without the relevant models, a user would not be able to playback the video file.
  • the models can be used to "lock" the video or be used as a key to access the video data.
  • the playback operation for the coded video data can depend on the models. This approach makes the encoded video data unreadable without access to the models.
  • This scheme can provide a user-friendly, developer-friendly, and efficient solution to restricting access to video content.
  • the models can progressively unlock the content.
  • an encoding might only decode to a certain level; then with progressively more complete models, the whole video would be unlocked.
  • Initial unlocking might enable thumbnails of the video to be unlocked, giving the user the capability of determining if they want the full video.
  • a user that wants a standard definition version would procure the next incremental version of the models.
  • the user needing high definition or cinema quality would download yet more complete versions of the models.
  • the models are coded in such a way as to facilitate a progressive realization of the video quality commensurate with encoding size and quality, without redundancy.
  • example embodiments of the invention may extend conventional encoding/decoding processes.
  • the present invention may be applied with flexible macroblock ordering (FMO) and scalable video coding (SVC), which are themselves extensions to the basic H.264 standard.
  • FMO flexible macroblock ordering
  • SVC scalable video coding
  • FMO allocates macrob locks in a coded frame to one of several types of slice groups.
  • the allocation is determined by a macroblock allocation map, and macroblocks within a slice group do not have to be contiguous.
  • FMO can be useful for error resilience, because slice groups are decoded independently: if one slice group is lost during transmission of the bitstream, the macroblocks in that slice group can be reconstructed from neighboring macroblocks in other slices.
  • feature-based compression can be integrated into the "foreground and background" macroblock allocation map type in an FMO implementation.
  • Macroblocks associated with features comprise foreground slice groups, and all other macroblocks (those not associated with features) comprise background slice groups.
  • SVC provides multiple encodings of video data at different bitrates.
  • a base layer is encoded at a low bitrate, and one or more enhancement layers are encoded at higher bitrates.
  • Decoding of the SVC bitstreams can involve just the base layer (for low bitrate/low quality applications) or some or all of the enhancement layers as well (for higher bitrate/quality applications).
  • the substreams of the SVC bitstream are themselves valid bitstreams, the use of SVC provides increased flexibility in different application scenarios, including decoding of the SVC bitstream by multiple devices (at different qualities, depending on device capabilities) and decoding in environments with varying channel throughput, such as Internet streaming.
  • feature-based compression can be integrated into a quality scalability implementation by including the primary feature-based predictions in the base layer (see the section above on model- based primary and secondary predictions).
  • the coded frames in the base layer can then serve as reference frames for coding in the enhancement layer, where secondary feature- based predictions can be used.
  • information from feature -based predictions can be added incrementally to the encoding, instead of all at once.
  • all feature-based predictions (primary and secondary) can be moved to enhancement layers, with only conventional predictions used in the base layer.

Landscapes

  • Engineering & Computer Science (AREA)
  • Multimedia (AREA)
  • Signal Processing (AREA)
  • Compression Or Coding Systems Of Tv Signals (AREA)

Abstract

Le codec de compression basé sur les modèles selon l'invention applique une modélisation de niveau supérieur pour produire des prédictions meilleures que celles qu'on peut trouver au moyen de l'estimation et la compensation de mouvement à base de bloc conventionnelles. Des algorithmes de détection d'attributs et objets basés sur la vision informatique identifient des zones d'intérêt à travers le cube de données vidéo. Les attributs et objets détectés sont modélisés avec un ensemble compact de paramètres, et des instances d'attribut/objet similaires sont associées à travers les trames. Les attributs/objets associés sont formés en pistes et attribués à des blocs spécifiques de données vidéo à encoder. Les informations de pistage sont utilisées pour produire des prédictions basées sur un modèle pour ces blocs de données, permettant une navigation plus efficace de l'espace de recherche de prédiction que ce que l'on réalise généralement au moyen des procédés d'estimation de mouvement conventionnels. Une armature hybride permet la modélisation de données à des fidélités multiples et sélectionne le niveau approprié de modélisation pour chaque partie de données vidéo.
EP13710636.5A 2012-03-26 2013-02-07 Encodage et décodage vidéo basés sur le contexte Withdrawn EP2815572A2 (fr)

Applications Claiming Priority (4)

Application Number Priority Date Filing Date Title
US201261615795P 2012-03-26 2012-03-26
US201261707650P 2012-09-28 2012-09-28
US13/725,940 US9578345B2 (en) 2005-03-31 2012-12-21 Model-based video encoding and decoding
PCT/US2013/025123 WO2013148002A2 (fr) 2012-03-26 2013-02-07 Encodage et décodage vidéo basés sur le contexte

Publications (1)

Publication Number Publication Date
EP2815572A2 true EP2815572A2 (fr) 2014-12-24

Family

ID=47901315

Family Applications (1)

Application Number Title Priority Date Filing Date
EP13710636.5A Withdrawn EP2815572A2 (fr) 2012-03-26 2013-02-07 Encodage et décodage vidéo basés sur le contexte

Country Status (5)

Country Link
EP (1) EP2815572A2 (fr)
JP (1) JP2015515806A (fr)
CA (1) CA2868448A1 (fr)
TW (1) TW201342926A (fr)
WO (1) WO2013148002A2 (fr)

Families Citing this family (24)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9578345B2 (en) 2005-03-31 2017-02-21 Euclid Discoveries, Llc Model-based video encoding and decoding
US9743078B2 (en) 2004-07-30 2017-08-22 Euclid Discoveries, Llc Standards-compliant model-based video encoding and decoding
US9532069B2 (en) 2004-07-30 2016-12-27 Euclid Discoveries, Llc Video compression repository and model reuse
EP3058736B1 (fr) 2013-10-14 2019-02-27 Microsoft Technology Licensing, LLC Options côté codeur pour mode d'intra-prédiction de copie de bloc pour codage vidéo et d'image
KR102257269B1 (ko) 2013-10-14 2021-05-26 마이크로소프트 테크놀로지 라이센싱, 엘엘씨 비디오 및 이미지 코딩 및 디코딩을 위한 인트라 블록 카피 예측 모드의 피쳐
CA2925183C (fr) 2013-10-14 2020-03-10 Microsoft Technology Licensing, Llc Fonctions de mode carte d'index de couleur de base pour codage et decodage de video et d'image
JP6355744B2 (ja) 2014-01-03 2018-07-11 マイクロソフト テクノロジー ライセンシング,エルエルシー ビデオ及び画像符号化/デコーディングにおけるブロックベクトル予測
US10390034B2 (en) 2014-01-03 2019-08-20 Microsoft Technology Licensing, Llc Innovations in block vector prediction and estimation of reconstructed sample values within an overlap area
US11284103B2 (en) 2014-01-17 2022-03-22 Microsoft Technology Licensing, Llc Intra block copy prediction with asymmetric partitions and encoder-side search patterns, search ranges and approaches to partitioning
US10542274B2 (en) 2014-02-21 2020-01-21 Microsoft Technology Licensing, Llc Dictionary encoding and decoding of screen content
US10097851B2 (en) 2014-03-10 2018-10-09 Euclid Discoveries, Llc Perceptual optimization for model-based video encoding
US10091507B2 (en) 2014-03-10 2018-10-02 Euclid Discoveries, Llc Perceptual optimization for model-based video encoding
WO2015138008A1 (fr) 2014-03-10 2015-09-17 Euclid Discoveries, Llc Suivi de bloc continu pour prédiction temporelle en codage vidéo
TWI499985B (zh) * 2014-04-30 2015-09-11 Univ Nat Taiwan Science Tech 影像特徵點自動點選方法和對應系統
EP3158734A4 (fr) 2014-06-19 2017-04-26 Microsoft Technology Licensing, LLC Modes de copie intra-bloc et de prédiction inter unifiés
US9767853B2 (en) 2014-07-21 2017-09-19 International Business Machines Corporation Touch screen video scrolling
CA2959682C (fr) 2014-09-30 2022-12-06 Microsoft Technology Licensing, Llc Regles pour modes de prediction intra-image lorsqu'un traitement parallele de fronts d'onde est active
WO2016197314A1 (fr) 2015-06-09 2016-12-15 Microsoft Technology Licensing, Llc Encodage/décodage robuste de pixels avec code d'échappement en mode palette
US10986349B2 (en) 2017-12-29 2021-04-20 Microsoft Technology Licensing, Llc Constraints on locations of reference blocks for intra block copy prediction
US11830225B2 (en) 2018-05-30 2023-11-28 Ati Technologies Ulc Graphics rendering with encoder feedback
CN110503636B (zh) * 2019-08-06 2024-01-26 腾讯医疗健康(深圳)有限公司 参数调整方法、病灶预测方法、参数调整装置及电子设备
CN111083497B (zh) * 2019-12-31 2022-07-08 北京奇艺世纪科技有限公司 一种视频截图方法及装置
CN113938666B (zh) * 2021-10-25 2023-07-25 珠海普罗米修斯视觉技术有限公司 基于关键帧的视频数据传输方法、装置及存储介质
US20240146963A1 (en) * 2022-10-17 2024-05-02 Alibaba Damo (Hangzhou) Technology Co., Ltd. Method and apparatus for talking face video compression

Family Cites Families (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6738424B1 (en) * 1999-12-27 2004-05-18 Objectvideo, Inc. Scene model generation from video for use in video processing
US7457435B2 (en) 2004-11-17 2008-11-25 Euclid Discoveries, Llc Apparatus and method for processing video data
EP2602742A1 (fr) 2004-07-30 2013-06-12 Euclid Discoveries, LLC Appareil et procédé de traitement de données vidéo
WO2010042486A1 (fr) * 2008-10-07 2010-04-15 Euclid Discoveries, Llc Compression vidéo basée sur une caractéristique
US7508990B2 (en) 2004-07-30 2009-03-24 Euclid Discoveries, Llc Apparatus and method for processing video data
US7457472B2 (en) 2005-03-31 2008-11-25 Euclid Discoveries, Llc Apparatus and method for processing video data
US7436981B2 (en) 2005-01-28 2008-10-14 Euclid Discoveries, Llc Apparatus and method for processing video data
CN101061489B (zh) 2004-09-21 2011-09-07 欧几里得发现有限责任公司 用来处理视频数据的装置和方法
KR101216161B1 (ko) * 2005-03-31 2012-12-27 유클리드 디스커버리스, 엘엘씨 비디오 데이터를 프로세싱하는 장치 및 방법
JP2010526455A (ja) * 2007-01-23 2010-07-29 ユークリッド・ディスカバリーズ・エルエルシー 画像データを処理するコンピュータ方法および装置
US8848802B2 (en) * 2009-09-04 2014-09-30 Stmicroelectronics International N.V. System and method for object based parametric video coding

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See references of WO2013148002A2 *

Also Published As

Publication number Publication date
JP2015515806A (ja) 2015-05-28
CA2868448A1 (fr) 2013-10-03
TW201342926A (zh) 2013-10-16
WO2013148002A3 (fr) 2013-12-19
WO2013148002A2 (fr) 2013-10-03

Similar Documents

Publication Publication Date Title
US9578345B2 (en) Model-based video encoding and decoding
US9743078B2 (en) Standards-compliant model-based video encoding and decoding
WO2013148002A2 (fr) Encodage et décodage vidéo basés sur le contexte
US20130114703A1 (en) Context Based Video Encoding and Decoding
US8902971B2 (en) Video compression repository and model reuse
US9532069B2 (en) Video compression repository and model reuse
US9621917B2 (en) Continuous block tracking for temporal prediction in video encoding
US10291925B2 (en) Techniques for hardware video encoding
CN111801945A (zh) 具有基于边信息的视频编译的混合运动补偿神经网络
US10574997B2 (en) Noise level control in video coding
CA2868784A1 (fr) Referentiel de compression video et modele de reutilisation
US10623744B2 (en) Scene based rate control for video compression and video streaming
US11025913B2 (en) Encoding video using palette prediction and intra-block copy
CN114175658A (zh) 用于改进视频编解码中的运动估计的方法和装置
CN116233434A (zh) 用于视频编码的变换块级扫描顺序选择
US8611415B1 (en) System and method for coding using improved motion estimation
KR20220032623A (ko) 복수의 컬러 성분들의 공동 변환 코딩
CN117501694A (zh) 用于视频编解码中的仿射运动补偿的具有双向匹配的运动细化
Wiegand et al. Entropy-constrained linear vector prediction for motion-compensated video coding
Chen et al. An improved 3D wavelet-based scalable video coding codec for MC-EZBC
US20230300341A1 (en) Predictive video coding employing virtual reference frames generated by direct mv projection (dmvp)
US20240073438A1 (en) Motion vector coding simplifications
US20240214565A1 (en) Method, device, and medium for video processing
US20230050102A1 (en) Triangulation-Based Adaptive Subsampling of Dense Motion Vector Fields
Karmakar et al. Efficient motion modelling with variable-sized blocks from hierarchical cuboidal partitioning

Legal Events

Date Code Title Description
PUAI Public reference made under article 153(3) epc to a published international application that has entered the european phase

Free format text: ORIGINAL CODE: 0009012

17P Request for examination filed

Effective date: 20140917

AK Designated contracting states

Kind code of ref document: A2

Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR

AX Request for extension of the european patent

Extension state: BA ME

DAX Request for extension of the european patent (deleted)
STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: THE APPLICATION HAS BEEN WITHDRAWN

18W Application withdrawn

Effective date: 20171114