EP1031238A4 - SYSTEM AND METHOD FOR COMPRESSING VIDEO DATA - Google Patents
SYSTEM AND METHOD FOR COMPRESSING VIDEO DATAInfo
- Publication number
- EP1031238A4 EP1031238A4 EP98958556A EP98958556A EP1031238A4 EP 1031238 A4 EP1031238 A4 EP 1031238A4 EP 98958556 A EP98958556 A EP 98958556A EP 98958556 A EP98958556 A EP 98958556A EP 1031238 A4 EP1031238 A4 EP 1031238A4
- Authority
- EP
- European Patent Office
- Prior art keywords
- subband
- data
- block
- recited
- macro
- 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- Withdrawn
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/90—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using coding techniques not provided for in groups H04N19/10-H04N19/85, e.g. fractals
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/48—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using compressed domain processing techniques other than decoding, e.g. modification of transform coefficients, variable length coding [VLC] data or run-length data
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/50—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding
- H04N19/503—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding involving temporal prediction
- H04N19/51—Motion estimation or motion compensation
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/60—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using transform coding
- H04N19/61—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using transform coding in combination with predictive coding
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/60—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using transform coding
- H04N19/61—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using transform coding in combination with predictive coding
- H04N19/619—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using transform coding in combination with predictive coding the transform being operated outside the prediction loop
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/60—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using transform coding
- H04N19/63—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using transform coding using sub-band based transform, e.g. wavelets
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/60—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using transform coding
- H04N19/63—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using transform coding using sub-band based transform, e.g. wavelets
- H04N19/64—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using transform coding using sub-band based transform, e.g. wavelets characterised by ordering of coefficients or of bits for transmission
- H04N19/645—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using transform coding using sub-band based transform, e.g. wavelets characterised by ordering of coefficients or of bits for transmission by grouping of coefficients into blocks after the transform
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
- H04N19/102—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or selection affected or controlled by the adaptive coding
- H04N19/115—Selection of the code volume for a coding unit prior to coding
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
- H04N19/134—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or criterion affecting or controlling the adaptive coding
- H04N19/146—Data rate or code amount at the encoder output
Definitions
- the present invention relates generally to apparatus and methods for encoding and decoding video information. More particularly, the present invention relates to an apparatus and method for motion estimation and motion prediction in the transform domain.
- Video encoding techniques attempt to encode video information with as few bits as possible, while still maintaining the image quality required for a given application.
- video compression techniques attempt to reduce the bandwidth required to transmit a video signal by removing redundant information and representing the remaining information with a minimum number of bits, from which an approximation to the original image can be reconstructed, with a minimal loss of important features. In this manner, the compressed data can be stored or transmitted in a more efficient manner than the original image data.
- There are a number of video encoding techniques which improve coding efficiency by removing statistical redundancy from video signals.
- DCT Discrete Cosine Transform
- MPEG video encoding technique for example, developed by the Motion Pictures Experts Group, achieves significant bit rate reductions by taking advantage of the correlation between pixels (pels) in the spatial domain (through the use of the DCT), and the correlation between image frames in the time domain (through the use of prediction and motion compensation).
- Transform encoders based on DCT block the image primarily for two reasons: 1) experience has shown that the DCT is a good approximation to the known optimal transform (Kahunen-Luove') on 8 x 8 regions of the image or a sequence of difference images; and 2) the processing of DCT grows O(N log N ) and through the blocking of the image, computational effort is limited.
- DCT based approaches unless otherwise enhanced, have basis functions which are compactly supported by (or zero outside of) an 8 x 8 region of an image.
- the orthogonal and bi-orthogonal transforms under consideration have basis members which are predominately supported in a finite interval of the image, but share extent with neighboring spatial regions.
- Subband image encoding techniques for example, divide an input image into a plurality of spatial frequency bands, using a set of filters and then quantize each band or channel. For a detailed discussion of subband image encoding techniques see Subband Video Coding With Dynamic Bit Allocation and Geometric Vector Quantization, C. Podilchuck & A. Jacquin, SPIE Vol. 1666
- FIG. 1 illustrates a simplified block diagram of a prior art standard video compression approach using DCT.
- the changes in the image sequence are efficiently represented through motion detection techniques such as one technique used in MPEG when in predictive mode.
- motion detection techniques such as one technique used in MPEG when in predictive mode.
- a previous frame is used as a reference frame and a subsequent frame, in forward prediction, is compared against the previous frame to eliminate temporal redundancies and rank the differences between them according to degree.
- This step sets the stage for motion prediction of the subsequent frame and also reduces the data size of the subsequent frame.
- a determination is made as to which parts of the image have moved.
- interframe motion prediction is carried out by applying motion compensation techniques to the reference frame and subsequent frame.
- the resulting prediction is subtracted from the subsequent frame to generate a prediction error/frame.
- the changes are converted to features.
- MPEG this is done by compressing the prediction error using a 2- dimensional 8 x 8 DCT.
- a "lossy" system refers to a system that loses precision through the various stages of the encoder and thereby lacks the ability to substantially reconstruct the input from the transform coefficients when decoding. The inability to compensate for the reduced accuracy exhibited by these low precision transforms have been an impediment to the use of such transforms.
- the subject invention is directed to a novel and unique apparatus and method for compressing data. More particularly, the present apparatus and method are adapted and configured to more efficiently encode data representing, for example, a video image, thereby reducing the amount of data that must be transferred to a decoder.
- the invention concerns a method of compressing data that includes a first data set and a second data set.
- the method includes transforming the first and second data sets into corresponding first and second transform coefficient sets. Thereafter, data is generated which represents differences between the first and second transform coefficient sets. The generated data is then encoded for transmission to the decoder.
- Transforming the first and second data sets may be performed utilizing a tensor product wavelet transform. Further, the remainders resulting from the transforming process may be transmitted from one subband to another subband.
- the data representing differences between the first and second transform coefficient sets is generated by estimating the differences between the first and second transform coefficient sets to provide motion vectors.
- the motion vectors are applied to the first transform coefficient set to produce a prediction of the second transform coefficient set.
- the prediction is subtracted from the second transform coefficient set resulting in a set of prediction errors.
- the first and second transform coefficient sets can be error corrected to ensure synchronization between the encoder and the decoder.
- a search region is generated about a subset of the transform coefficients from one of the first and the second transform coefficient sets. Thereafter, a related subset of transform coefficients is applied from the other of the first and the second transform coefficient sets to the search region. Then, the related subset of transform coefficients is traversed incrementally within the search region to a position representing a best incremental match. The related subset can then be traversed fractionally within the search region to a position representing a best fractional match.
- Another embodiment of the method of compressing data that includes a first data set and a second data set includes transforming the first and second data sets into corresponding first and second collections of subbands. Then, generating data representing the differences between the first and second collections of subbands.
- the data may be generated, for example, by carrying out a motion compensation technique.
- the motion compensation technique may provide output such as motion vectors and prediction errors. Thereafter, the generated data is encoded for transmission to the decoder.
- An embodiment may also the second collection of subbands macro-block packed to form a subband macro-block grouping. Thereafter, the generated data may be obtained through a motion compensation technique as follows. The differences between the first collection of subbands and the subband macro-block grouping is estimated to provide motion vectors. The motion vectors are applied to the first collection of subbands producing a prediction of the second collection of subbands. The prediction is then subtracted from the second collection of subbands resulting in a set of prediction errors.
- the differences can be estimated between the first collection of subbands and the subband macro-block grouping as follows.
- a search region is generated about a subset of transform coefficients from the first collection of subbands.
- a related subset of transform coefficients from the subband macro-block grouping is applied to the search region.
- the related subset of transform coefficients is then traversed incrementally within the search region to a position representing a best incremental match.
- the related subset of transform coefficients is traversed fractionally within the search region to a position representing a best fractional match.
- a subband macro-block packing method for organizing subband blocks of a collection of subbands derived from a transform of an image.
- the method includes disassociating a set of related subband blocks from a collection of subbands that correspond to an image macro-block in the image.
- the set of related subband blocks are packed together as a subband macro-block.
- the steps of the disassociating and packing related subband blocks are repeated for each set of related subband blocks in the collection of subbands to form a subband macro-block grouping.
- the method for macro-block packing may be further refined by arranging the set of related subband blocks within the subband macro-block in the same relative position the subband blocks occupy in the collection of subbands.
- the method may also include locating the subband macro-block within the subband macro-block grouping in the same spatial location as the corresponding image macro-block is located within the image macro-block grouping.
- G - transform coefficients of the second subband macro- block grouping; and R reference (e.g., first subband macro-block grouping).
- the embodiment may include determining a low pass component and a high pass component of an image.
- the low pass component is normalized to generate a low pass normalized output and a first remainder (rl).
- the high pass component is normalized to generate a high pass normalized output and a second remainder (rh).
- a first operation (g(rl,rh)) is performed on the first and second remainders (rl, rh) and added to the results emanating therefrom to the approximation.
- a second operation (f(rl,rh)) is also performed on the first and second remainders (rl, rh) and added to the results emanating therefrom to the detail. It is important to note that the propagation of the remainders (propagation of errors) can be used in any transform, not just the tensor product.
- the above finite precision method results in an overcomplete representation of an image.
- the method may include downsampling, for example, by two (2), of the high and low pass components to obtain the necessary and sufficient transform coefficients representing the image in the transform domain.
- An embodiment of the finite precision method includes a low pass filter having the value -1, 2, 6, 2, -1 and a high pass filter having the value -1, 2, -1.
- the first operation (g(rl,rh)) and the second operation (f(rl,rh)) have the functions:
- a particular example of a tensor product wavelet transform including the above has the form:
- X ll ⁇ data that precedes input data X 2l ;
- A, approximation term (decimated low pass filter output). Also disclosed is an encoder apparatus for predicting changes between a sequence of frames in the transform domain.
- the apparatus includes a transformation device, having an input configured to receive a first and second frame of the sequence of frames, and further configured to generate therefrom a corresponding first and second collection of subbands that each support a set of transform coefficients.
- a motion compensation device having an input coupled to the transformation device, is configured to receive the first and second collection of subbands, and further configured to efficiently represent differences between the first and second collection of subbands.
- a difference block having an input coupled to the transformation device and an input coupled to the output of the motion compensation device. The input received from the motion compensation device is subtracted from the second collection of subbands in the difference block, thereby generating a prediction errors.
- the motion compensation device includes a motion estimation device configured to compare the first and second collection of subbands. A collection of motion vectors is generated therefrom which approximately represent the differences between the first and second collections of subbands.
- the motion compensation device also includes a motion prediction device, having an input coupled to the motion estimation device, configured to receive the motion vectors and the first collection of subbands, and further configured to generate therefrom a prediction grouping representing a prediction of the second collection of subbands. The prediction of the second collection of subbands is subtracted from the second collection of subbands in a difference block resulting in prediction errors.
- a finite precision transforming apparatus for transforming an image frame into the transform domain.
- the apparatus includes a low pass component and a high pass component arranged in parallel and sharing an input that is configured to receive the image frame.
- a low pass normalizing device is included which has an input configured to receive the output of the low pass component and is further configured to produce a low pass normalized output and a first remainder (rl).
- a high pass normalizing device has an input configured to receive the output of the high pass component and is further configured to produce a high pass normalized output and a second remainder (rh).
- a first operation device has an input configured to receive the first remainder (rl) and the second remainder (rh) and further configured to calculate a first calculation (g(rl,rh)) thereby generating a first calculation result.
- a second operation device has an input configured to receive the first remainder (rl) and the second remainder (rh) and configured to calculate a second calculation (f(rl,rh)) thereby generating a second calculation result.
- a first adder has an input configured to receive the low pass normalized output and the first calculation result, the first adder generating a subband approximation.
- a second adder has an input configured to receive the high pass normalized output and the second calculation result, the second adder generating a subband detail.
- the finite precision transforming apparatus further includes a first downsampler at the low pass output and a second downsampler at the high pass output. A downsampling of two (2) provides sufficient and necessary transform coefficients to reconstruct the input image in the decoder.
- FIG. 1 is a schematic block diagram of a prior art standard video compression approach using the Discrete Cosine Transform (DCT) wherein motion compensation is carried out in the image domain;
- DCT Discrete Cosine Transform
- FIG. 2 is a schematic block diagram illustrating a general arrangement of an embodiment of the present invention including provisions for motion compensation to be carried out in the transform domain;
- FIG. 3 is a schematic block diagram of a more detailed arrangement of the embodiment illustrated in FIG. 2;
- FIG. 4(a) illustrates a QCIF image having image macro-blocks (IMB X x ) 0,0 through 8,10 and FIG. 4(b) illustrates a subband representation of the QCIF image after the image frame has been transformed by a forward wavelet transform;
- FIG. 5(a) illustrates the subband representation of the QCIF image as illustrated in FIG. 4(b)
- FIG. 5(b) illustrates a collection of subband macro-blocks (SMB X x ) generated from the subband representation illustrated in FIG. 5(a)
- FIG. 5(c) illustrates the organization of the subband macro-blocks of FIG. 5(b) so that the subband macro-blocks (SMB X x ) correspond spatially with their related image macro- blocks (IMB x?x ) of FIG. 4(a);
- FIGS. 6(a) and 6(b) are schematic block diagrams illustrating filter banks for transforming and decimating an input image and their respective vertical and horizontal subbands created from each filter bank;
- FIG. 7 illustrates an architecture for transferring the finite precision arithmetic in the filter banks from the high band region to the low band region and, conversely, from the low band region to the high band region;
- FIG. 8 illustrates search regions in the transform domain for each subband (SB/,-) corresponding to image macro-block 2,4 (IMB2 ; 4) in the image domain wherein the search band is P x P pels, and further details the search region for that of SBQO when the input image size is in QCIF;
- FIGS. 9(a) through 9(d) illustrate a method by which motion is estimated in the transform domain
- FIG. 10 illustrates a method by which motion is predicted in the transform domain
- FIG. 11 is a schematic block diagram illustrating another detailed arrangement of the embodiment illustrated in FIG. 2;
- FIG. 12 is a schematic block diagram illustrating another detailed embodiment of the invention wherein motion estimation is carried out in the image domain and motion prediction is carried out in the transform domain;
- FIG. 13 illustrates a P x P pels search region when searching in the image domain about image macro-block 2,4 (IMB2 4) when the input size is QCIF;
- FIG. 14 is a schematic block diagram illustrating another detailed embodiment of the invention wherein motion estimation and motion prediction is carried out in the image domain.
- An embodiment of the present invention provides an apparatus and method for compressing digital video signals using a limited precision transformation technique.
- the embodiment improves on conventional loss-less or lossy transform based techniques by motion compensating, e.g., estimating and predicting motion, in the transform domain, rather than in the time domain as in the prior art. In this manner, improved image quality can be achieved on less expensive hardware.
- motion compensation is intended to be defined in its broadest sense. In other words, although motion compensation is often described and is illustrated herein as including motion estimation and motion prediction of a group of picture elements, it should also be understood to encompass, for example, rotation and scale. In addition, the term “motion compensation” may include, for example, simply generating data representing differences between two sets of data.
- An image sequence can be readily understood to be a collection of spatially oriented data elements (either scalar, vector, or functional) which are placed in arrangement with each other and are indexed by time or some other parameter.
- An image sequence can be in Cartesian coordinates, but other coordinate systems in the art can be used.
- the present apparatus and method can be utilized in non- video applications such as speech, audio, and electrocardiogram compression. That is, even though the invention disclosed herein is illustrated on a two-dimensional system (2 D), i.e., video compression, it is intended that the teachings can be applied to any other dimensional systems so to advance the art of data compression in general.
- 2 D two-dimensional system
- teachings can be applied to one and one-half dimensional systems (1-1/2 D) such as ultra-sound imaging. Also, the teachings can be applied to three dimensional systems (3 D) such as magnetic resonance imaging (MRI).
- 1-1/2 D such as ultra-sound imaging
- 3 D three dimensional systems
- MRI magnetic resonance imaging
- the term "frame” refers to a single image of a sequence of images fed to an encoder, regardless of the form of the single image, i.e., regardless if it is in the time domain, the frequency domain, or of any other processing that has been done on it.
- the term "pel” is used in reference to a picture element in the time domain and the terms “coefficient” and “transform coefficient” are used in reference to representations of the pels which are generated after the pels have passed through, for example, a forward wavelet transform. These terms are used to facilitate the description of the embodiments and are in no way intended to restrict the scope of the invention.
- FIG. 2 a schematic block diagram of an embodiment for compressing a sequence of images or sequence of frames. This diagram is one of several embodiments disclosed herein. More detailed embodiments are discussed in the paragraphs that follow.
- an image is converted to a collection of features in the transform domain in block 20.
- the features that are determined to be significant for that image, i.e., those features that are determined to have significantly changed from a past or reference frame, are selected in block 22.
- the significant features are efficiently represented in block 24 and, thereafter, sent to a decoder to update the features in a reference frame.
- the original image is transformed in block 20 and represented by a transform coefficient set.
- the transform coefficients of the coefficient set are then evaluated in block 22 to determine their significance via various weighting and evaluation techniques and ranked according the their significance.
- motion compensation may include motion estimating the change between frames to generate a set of motion vectors.
- the motion vectors are applied to a reference frame during a motion prediction step.
- the results from motion prediction are subtracted from the transform coefficient set to determine the errors of that prediction.
- the prediction errors are then optionally scaled and finally positionally encoded along with the motion vectors for transmittal to the decoder. Referring to FIG.
- FIG. 3 a schematic block diagram illustrates a more particular arrangement of the embodiment that was described with reference to FIG. 2.
- An image sequence or series of video frames 26 encoded in, for example, Caltech Intermediate Format (CIF) are fed to a converter 28.
- a CIF frame has 288 x 352 pels.
- the frames are converted to quarter CIF (QCIF), for example, QCIF image 30 as illustrated in FIG. 4a.
- QCIF image has 144 x 176 pels.
- CIF is converted to QCIF by low-pass filtering and decimating by two (2) in both the horizontal and vertical directions.
- the 144 x 176 pels are divided into image macro- blocks (IMB x x ), each having 16 x 16 pels.
- QCIF is used herein as an example only and is not intended in any way as a limitation with respect to this invention.
- the techniques described below are readily adaptable to other image (and non-image) formats by methods well known to those skilled in the art
- QCIF image 30 is fed to blocks 32 and 36, which make up block 20 of FIG. 2, wherein mapping of the images to features takes place. More specifically, QCIF image 30 (FIG. 4(a)) is fed to block 32 wherein a forward wavelet transform transforms each frame to a collection of subbands 34 (FIG. 4(b)).
- This organization of the transformed image, i.e., collection of subbands 34 is stored in memory for later use, for example, for motion estimation, motion prediction, and determining prediction error.
- An appropriate forward wavelet transform that may be used for this invention is discussed in greater detail herein below.
- the collection of subbands 34 are fed to block 36 for subband macro-block packing.
- subband macro-block grouping 40 The collection of all subband macro-blocks for a particular frame is called a subband macro-block grouping 40.
- FIG. 5 illustrates the process of subband macro-block packing.
- all related subband blocks in the collection of subbands 34 (FIG 5(a)) are reorganized during subband macro-block packing to form subband macro- blocks 38 as illustrated in FIG 5(b).
- subband macro-block SMB2 4 As illustrated FIG. 5(b), the shaded subband blocks in FIG. 5(a), corresponding to image macro-block 2,4 (IMB2 4) in FIG. 4(a), are reorganized during subband macro-block packing in block 36 (FIG. 3) to form subband macro-block SMB2 4 as illustrated FIG. 5(b).
- the subband macro-blocks 38 (SMBQ 0 through SMBg ⁇ Q) are then organized into subband macro-block grouping 40 as illustrated in FIG. 5(c) such that each subband macro-block is supported by the spatial position of its corresponding image macro-block (IMB X x ) in QCIF image 30.
- SMB2 4 is found to be significantly supported by the spatial location of IMB2 4 as illustrated in FIGS. 4(a) and 5(c).
- SBQO subband 00
- SB33 subband block 118 in subband 33
- SB2 4 subband macro-block 2,4
- subband macro-block grouping 40 is fed to blocks 42, 46, 48, and 52, which make up block 22 found in FIG. 2, where it is determined which features or subband macro-blocks (SMBQ 0 through SMBg ⁇ Q) have changed.
- subband macro-block grouping 40 is fed to block 42 wherein weights are applied to scale each subband macro-block in subband macro-block grouping 40 by an amount which equalizes the perceptual importance of the subband macro-block.
- the output of weighting block 42 is weighted grouping 44.
- Perceptual importance through weighting can be determined, for example, through a Mean Opinion Score study, or determined from weights used in other coding systems such as those found in H.261 and H.263 of the Consultative Committee for International Chat and Telephone (CCITT), the standards of which are incorporated by reference herein.
- CITT Consultative Committee for International Circuit and Telephone
- Mean Opinion Scoring see Discrete Cosine Transform, K. R. Rao & P. Yip, Academic Press, Inc., pp. 165-74 (1990), incorporated by reference herein.
- weighted grouping 44 is fed to and processed in change detect block 46 to determine the relative amount of change that has occurred.
- This change is also termed the 'significance' or, for the purpose of video, the distortion of weighted grouping 44.
- Significance can be determined in relation to a given reference such as, for example, zero (0) or a past weighted grouping.
- the loop extending from change-detect block 46 includes a frame delay 48 which returns a past weighted grouping to change-detect block 46 for use as a reference.
- the output of change-detect block 46 is change detected grouping 50.
- a zero (0) reference is used in change detect block 46, for example, when initially transmitting frames through the encoder. In this case, the entire frame is referenced to zero (0). This is also known as mtr ⁇ frame referencing. As described above, a past weighted grouping can also be used wherein the macro-block grouping is weighted in block 42 as described above and thereafter delayed in delay block 48 of change-detect block 46 for use as a reference. This later method, also known as terframe referencing, eliminates repeatedly sending redundant and/or unimportant information to the decoder. An alternative use of zero (0) frame referencing is for reproducing and maintaining a relatively accurate reference image at the decoder during system operation.
- One method employs periodically applying a zero (0) reference to the entirety of every eighth (8th) frame of the standard 30 frames/second.
- the image can be stoichastically refreshed such as by randomly, or methodically, referencing subband blocks to zero (0).
- the zero-referenced subband blocks are identified as such so to prevent motion compensation operations (described below) to be performed on the effected blocks.
- the identified subband blocks are reproduced in whole at the decoder for refreshing either the entire reference or a part of the reference therein, as the case may be.
- the collection of subbands 34 that were earlier stored in memory and the subband macro-blocks of change-detected grouping 50 are ranked in block 52 according to the amount each subband block is determined to have changed, i.e., in accord with their significance. Ranking is based on the values previously assigned by weighting and detecting the subband macro-blocks in blocks 42 and 46, respectively.
- the output of block 52 includes a ranked subband grouping 53 and a ranked subband macro-block grouping 54 which are transmitted over line 55.
- ranked subband grouping 53 and ranked subband macro-block grouping 54 are selectively fed to blocks 56, 60, 62, 68, 72, and 76, which correspond to block 24 of FIG.
- ranked subband macro-block grouping 54 (the 'present' frame) is fed to block 56 for motion estimation.
- Ranked subband grouping 53 is fed to delay block 62, thereafter providing a delayed ranked subband grouping 57 (the 'reference' frame) to line 64 for motion estimation and motion prediction in blocks 56 and 60, respectively.
- a collection of motion vectors 58 are generated in motion estimation block 56, in a manner to be described below, and fed to block 60 for motion prediction and also sent to block 76 for positional encoding.
- the motion vectors 58 sent to motion prediction block 60 are used to alter delayed ranked subband grouping 57 so to generate a predicted grouping 66.
- a difference block 68 receives ranked subband grouping 53 and subtracts predicted grouping 66 therefrom, resulting in grouping differences 70, i.e., the prediction error.
- the grouping differences 70 are further scaled in block 72 resulting in scaled grouping differences 74.
- the scaled grouping differences 74 from scaling block 72 and the collection of motion vectors 58 from motion estimation block 56 are positionally encoded as macro- blocks in block 76. Therein, the data is efficiently organized into a bit stream. Encoded bit stream grouping 78 is output from block 76 and transmitted via transmission line 80 to a decoder 82 for inverse processing. Transmission can be through a variety of mediums, for example, electronic, electromagnetic, or optical.
- bit stream formatting there are several standard methods well known in the art for formatting bit streams.
- the format used in an H.263 based encoder system is one example.
- a bit stream is basically a serial string of bit packets. Each packet representing a particular category of data.
- bit packets may include system level data, video, control, and audio data.
- data is received for positional encoding in block 76, it is organized into bit packets in accordance with the format in use.
- bit packets representing a video frame starts with a bit identifying it as a new frame.
- the amount of quantization and other control codes typically follow.
- macro-blocks representing the scaled grouping differences 74.
- the number of macro-blocks equals ninety-nine (99). (See FIG.
- each macro-block is proceeded by a macro-block zero bit (MBZero-bit) which indicates the presence or absence of non-zero data in a macro-block.
- MZero-bit macro-block zero bit
- control information for the macro- block including the related collection of motion vectors 58, is sent followed by the subband data, i.e., the related scaled grouping differences 74.
- Including such information substantially reduces the number of bits that are sent over transmission line 80 in that the absence of a macro-block is represented by a single symbol instead of the all of the bits that would be necessary to identify the entire string of macro-block coefficients that are equal to zero.
- An embodiment includes the step of flagging the subbands whose coefficients are equal to zero with a subband zero flag (SBZero flag).
- SBZero flag A subband from scaled grouping differences 74 whose coefficients are zero indicates that no changes were found to exist between corresponding subband blocks of ranked subband grouping 53 and predicted grouping 66. It takes substantially fewer bits to represent SBZero flags than to separately represent each coefficient equaling zero.
- the decoder is programmed to recognize both the MBZero-bit and the SBZero flag so to interpret the symbol introduced during positional encoding in block 76.
- An example of zero-runs length codes for symbolizing strings of zeros follows.
- encoded bit stream grouping 78 is received by decoder 82 via transmission line 80 and is fed to a positional decoding block 86 which reverses the effect of positional encoding block 76.
- the collection of motion vectors 58 are extracted from bit stream grouping 78 and fed to a prediction block 98.
- the decoded scaled grouping differences 88 in subband form (FIG. 4(b)), are provided to a quantum recovery block 90.
- quantum recovery block 90 past transform coefficients, and past and present dequantization terms are used to recover the quantized transform coefficients' values, i.e., they are used to recreate the grouping differences 70.
- a collection of subbands 92, the encoder's reference frame, is fed to a delay block 94.
- a delayed collection of subbands 96 is fed from the delay block 94 to a prediction block 98.
- the collection of motion vectors 58 are applied to the delayed collection of subbands 96 in prediction block 98.
- the delayed collection of subbands 96 is altered to generate a predicted grouping 100, i.e., a subband representation of the updated image not including the grouping differences 70.
- Grouping differences 70 and predicted grouping 100 are added in an adder block 102 generating the collection of subbands 92, i.e., a new reference frame.
- an inverse wavelet transform is performed in block 104 on the collection of subbands 92. This step is essentially the reverse of the forward wavelet transform 32 that was briefly described above and which will be described in greater detail herein below.
- the resulting output from block 104 is a reconstructed image 105.
- transform block 32 utilizes a tensor product wavelet transform.
- tensor product wavelet transforms see Standard Wavelet Basis Compression of Images, Joel Rosiene and Ian Greenshields, Optical Engineering, Vol. 33, Number 8 (March 1994), incorporated by reference herein.
- Other finite precision transforms may be utilized such as the well-known Mallat, GenLOT, or Harr transforms.
- suitable alternative wavelet transforms see Wavelets and Filter Banks, G. Strang and T. Nguyen, Wellesley-Cambridge Press (1997), incorporated by reference herein.
- the forward wavelet transform process utilizes the tensor product wavelet transform or other well known finite precision transforms as modified herein to reduce the effects of a finite precision implementation.
- the transform process will consist of m x n stages to produce (m+1) x ⁇ n+ ⁇ ) subbands.
- the transform process consists of 3 x 3 stages to produce a total of sixteen (16) subbands.
- Other embodiments can be made following the disclosure provided herein that would be within the scope of this invention. Referring to FIG.
- a forward wavelet transform process initially filters a QCIF image frame 30 on a row-by-row basis using three stages.
- Each stage includes a low pass filter 108 and a high pass filter 110.
- each low pass filter 108 has a value of -1, 2, 6, 2, -1
- each high pass filter has a value of -1, 2, -1.
- the low pass components and high pass components are scaled and decimated, or downsampled, at each stage by decimators 112 and 114, respectively, whereby components of the sample values comprising a discrete signal are eliminated.
- the input image is downsampled by a factor of two (2) so to discard every other sample.
- the output of the first stage includes a low pass filter component A 0 R and a high pass component D 0 R.
- Low pass component A 0 R is decomposed a second time and then a third time resulting in additional row details D jR and D 2 R, and row average A 2R
- the row outputs D 0 R, D 1 R , D 2 R, and A 2 R of the row stages shown in FIG. 6(a) are then applied on a column-by-column basis to the stages shown in FIG. 6(b).
- Each of the three stages shown in FIG. 6(b) include a filter pair, downsampling, and normalization processes that are applied in the same manner as discussed above in conjunction with FIG. 6(a).
- the transform output is a collection of subbands 34 as discussed above with regard to FIG. 3 and as illustrated in FIG. 4(b).
- the shaded subband blocks for example, subband block 116 in SBQO and subband block 118 in SB33, correspond with IMB2 4 in QCIF image 30 of FIG. 4(a). Due to the decimation process described above, each corresponding subband block is reduced proportionally such that, for example, subband block 116 in SBQO includes 8 x 8 coefficients and subband block 118 in SB33 includes 2 x 2 coefficients.
- the related subband blocks e.g., those subband blocks in each subband (SBQO through SB33) that are found in subband positions 2,4, are collected during the step of subband macro-block packing in block 36 (FIGS. 3 and 5) to facilitate certain processing steps.
- the remainder for each stage of the subband encoding process is propagated to the opposite filter path in order to compensate for errors introduced due to the finite precision transform.
- the propagated remainder is utilized to adjust the coefficients on the opposite filter path to account for the loss of precision.
- the process results in a non-linear transform. Further, the process by which the filters are altered may make them neither bi-orthogonal nor orthogonal.
- FIG. 7 illustrates an implementation for propagating the remainders to opposite filter channels for the first stage of the row transform shown in FIG. 6(a).
- a similar implementation is included at each of the row stages and column stages.
- the coefficients of input frame 30 are filtered in low pass filter 108 and high pass filter 110 in the normal manner.
- the results are, respectively, downsampled in samplers 112 and 114.
- the decomposed results of low pass filter 108 are normalized in a low pass normalization process 120 producing a low pass normalized output 122 and a low pass remainder rl.
- the decomposed results of high pass filter 110 are normalized in a high pass normalization process 124 producing a high pass normalized output 126 and a high pass remainder rh.
- the remainders rl and rh resulting from each normalization process 120 and 124, respectively, are each passed through functions g(rl,rh) 128 and f(rl,rh) 130 as illustrated.
- the results of function g(rl,rh) 128 are added to low pass normalized output 122 in adder 132 resulting in A Q R (first stage averages).
- the results of function f(rl,rh) 130 are added to high pass normalized output 126 in adder 133 resulting in D 0 R (first stage lost details).
- X 2l ⁇ data that precedes input data X 2l ;
- the above description of the tensor product wavelet transform illustrates a two- way split into high pass (details) and low pass (approximations) components.
- the description illustrates the possibility of propagating remainders from a first band to a second band, from the second band to the first band, or both from the first band to the second band and from the second band to the first.
- the embodiment described above is intended to illustrate the basic concepts of the invention and should in no way be interpreted to limit the scope of the invention.
- a tensor product wavelet transform can have a first stage where a three-way split includes a high pass filter, a medium pass filter, and a low pass filter.
- the output of the low pass filter can then be iterated, i.e., a second stage having a three- way split can be applied to the output of the low pass filter, resulting in a total of five (5) subbands.
- the remainders could be propagated from the low pass filter and the high pass filter to the medium pass filter.
- the input can be split at each stage and interated, and also that there are numerous other ways in which the remainders can be propagated between subbands.
- the above description of the propagation of remainders is not intended to limit its use to a tensor product wavelet transform. It can be used with any transform.
- the propagation of remainders can be used with a Discrete Cosine Transform (DCT).
- DCT Discrete Cosine Transform
- the propagation of remainders can be used in a lossless or lossy manner.
- the output of forward wavelet transform 32 can be a complete representation or an over-complete representation of QCIF image 30.
- a complete representation of QCIF image 30 includes a collection of subbands that are just enough to represent the contents of the image.
- An over-complete representation of QCIF image 30 includes the complete representation and redundant, alternative, or additional subband representations to facilitate motion compensation as will be described herein below.
- Each representation has value in the disclosed embodiment.
- the over-complete representation can include a variety of image changes, such as translational movement, rotational movement, and scaling. These changes can be recalled as necessary during motion compensation, reducing the problem of representing image changes to one of indexing.
- change-detect block 46 includes a change detection metric, to which weighted grouping 44 is to be applied, of the general form:
- e c a measurement of distortion relative to reference R
- G a present grouping of subband transform coefficients
- R a reference such as, for example, zero (0) or a previous grouping of subband coefficients obtained through delay block 48.
- a change detection metric may take the more specific form:
- change-detect 46 can take advantage of information provided by a feed-back 132 (FIG. 3) from encoded bit stream grouping 78 to eliminate certain weighted macro-blocks in weighted grouping 44 if any are determined to be too expensive, in terms of bit allocation, to be output from change-detect 46.
- change-detect block 46 may replace one feature, e.g., subband block, with another which it deems better representative of the feature.
- ranked subband grouping 53 and ranked subband macro-block grouping 54 are fed to delay block 62 and block
- a comparison process is carried out between the subband blocks of ranked subband macro-block grouping 54, i.e., the 'present' frame, and related search regions of delayed ranked subband grouping
- the comparison process carried out in block 56 results in a collection of motion vectors 58 which are fed to block 60 for motion prediction and to block 76 for positional encoding into a bit stream, as briefly described herein above.
- motion estimation in block 56 and the generation of a collection of motion vectors 58 will now be more particularly described.
- delayed ranked subband grouping 57 is illustrated.
- Delayed ranked subband grouping 57 is similar to the collection of subbands 34 illustrated in FIG. 4(b), but has been further processed by having its subband blocks ranked in block 52 (FIG. 3) and by having been delayed by at least one frame in delay block 62.
- search regions are defined about subband blocks in at least one of the subbands (SBQO through SB33).
- the subband blocks within each subband that are selected to have search regions defined about them are those that were defined as significant in change-detect block 46. It is often sufficient to develop the motion vectors based on the significant subband blocks within SBQO-
- search regions that have been developed about each subband block that corresponds to image macro-block 2,4 (IMB2 4) of QCIF image 30 (FIG. 4(a)).
- the size of the search regions may be varied.
- the search regions about the subband blocks will always be proportional according to their fractional relationship with the image. For example, a basic search region of P x P pels in QCIF image 30 (FIG. 13) translates to a search region about subband block 137 in SBQO of P/2 X P/2 (FIG. 8), as indicated at 136, and a search region about subband block 140 in SBQI of P/4 x P/2 as indicated at 139.
- the P x P search region 107 of FIG. 13 is to include 32 x 32 pels, which is four (4) times the size of IMB2 4, having 16 x 16 pels. Therefore, the P/2 x P/2 search region 136 (FIG. 8) includes 16 x 16 coefficients which is four (4) times the size of subband block 137 (8 x 8 coefficients). And, the P/4 x P/2 search region 139 includes 16 x 8 coefficients which is four (4) times the size of subband block 140 (8 x 4 coefficients). As will be further described herein below, the subband search regions are used to facilitate determining motion vectors for each significant subband block (0,0 through 8,10) in some or all of the subbands (SBQO through SB33).
- the basic size (P x P) of the search region can be determined by empirical or statistical analysis taking into consideration, for example, the amount of movement anticipated between frames. In addition, consideration should be given to the computational effort needed to carry out a search in a given search region. It is readily understood by those skilled in the art that larger search regions require more computational resources and, hence, more interframe delay for a fixed processor. Conversely, smaller search regions require less computational resources but sacrifice image quality. This is especially true during high image-movement periods. That is, the quality of the image is reduced since part of the motion may be located out of the search region, thus preventing accurate motion vector selection.
- ranked subband grouping 53 and ranked subband macro- block grouping 54 are fed from block 52 to delay block 62 and motion estimation block 56, respectively, over line 55.
- a search region is placed about subband block 2,4 of SBoo in delayed ranked subband grouping 57 (FIG. 8).
- a subband block of SBQO i subband macro-block 2,4 in ranked subband macro- block grouping 54 (ref. subband block 116 in FIG. 5(c)) is used to traverse the search region for change.
- any selection of the subbands or all of the subbands may be used following the below described method.
- ranked subband grouping 53 is delayed in delay 62 producing delayed ranked subband grouping 57 (the 'reference' frame). Delayed ranked subband grouping 57 is fed to motion estimation block 56 wherein a search region 136 is identified to as having a P/2 x P/2 region in SBQO about subband block 137. For this example, the search region is equal to 16 x 16 coefficients.
- Ranked subband macro-block grouping 54 (the 'present' frame) is also fed to motion estimation block 56 wherein a subband block 138 (FIG. 9(a)), similar to the shaded area of subband block 116 in FIG. 5(c), is retrieved for use in the comparison process described below.
- FIGS. 9(a) through (d) there is illustrated the process by which motion vectors (MV X x ) are determined in motion estimation block 56 of FIG. 3.
- a motion vector is determined for one subband block, i.e., subband block 2,4, of SBQO-
- motion vectors can be determined for each significant subband block in each subband (SBQO -through SB33).
- subband block 138 of ranked subband macro-block grouping 54 is located within the search region 136 of delayed ranked subband grouping 57 (FIG. 8). Subband block 138 is essentially superimposed on subband block 137 of delayed ranked subband grouping 57.
- ranked subband macro-block grouping 54 has a structure similar to subband macro-block grouping 40 illustrated in FIG. 5(c).
- delayed ranked subband grouping 57 has a structure similar to the collection of subbands 34 illustrated in FIG. 4(b).
- coefficients 141 of search region 136 (illustrated as four (4) circles with an 'x' in each) and coefficients 142 of subband block 138 (illustrated as four (4) circles) are used herein to facilitate illustrating the method of determining motion vectors. It is assumed for this example that coefficients 141 and 142 are approximately equal in value and that the remaining coefficients (not shown) are of a different value from coefficients 141 and 142, but approximately equal to each other. The difference in the positions of coefficients 141 and 142 represents a change between two video frames, e.g., translational movement.
- subband block 138 traverses, i.e., searches in a predetermined step wise pattern, search region 136 seeking to determine the total absolute difference at each step between subband block 138 and search region 136.
- search region 136 seeking to determine the total absolute difference at each step between subband block 138 and search region 136.
- criterion other than total absolute difference can be used as a basis for the comparison.
- the initial comparison seeks to find the best match utilizing incremental, or whole step, movements of subband block 138.
- An incremental movement is a full shift, or step, in either the x or y direction.
- subband block 138 shifts within search region 136 by ⁇ 4 increments, i.e., transform coefficients, in the JC direction and ⁇ 4 increments in the y direction.
- Subband block 138 shifts ⁇ 4 increments in the x and y direction because subband block 138 has 8 x 8 coefficients while search region 136 has 16 x 16 coefficients.
- the best match is found to be three (3) full incremental movements in the positive x direction and two (2) full incremental movements in the positive y direction.
- fractional differences are determined to more accurately represent the difference between subband block 138 and search region 136.
- masks representing fractional movement appropriate for the particular subband are applied to subband block 138.
- SBQO is one-quarter (1/4) the size of the related macro- block in the original image (see IMB2 4 of FIG. 4(a))
- the four masks 143 are applied to subband block 138. Between the application of each mask the total absolute difference between the coefficients in subband block 138 and search region 136 is determined. If a better match is found, in comparison to that determined during incremental searching described above, the fractional mask is added to the motion vector. In the example, the best match is determined to be +1/2 fractional movement in the positive x direction. The resulting x and y components of the motion vector are +3 1/2 and +2, respectively.
- the 'best match' between the coefficients of the subband block and the coefficients of the search region may more accurately be described as the 'closest approximation' between the two. Motion prediction is used later to compensate for this lack of accuracy.
- the x and v components of the motion vector have their signs inversed and are scaled. More particularly, multiplying each of the x andy components by negative one (-1) and, in this example where SBQO is used for motion estimation, multiplying each of the x and y components by two (2).
- the signs of the x and y components are inversed so that when the motion vectors are applied to delayed ranked subband grouping 57 during motion prediction (discussed in more detail below), the appropriate coefficients are moved from the 'previous' frame position to the 'present' frame position.
- MV2 4 is stored in memory with the collection of motion vectors 58.
- MV2 4 therefore represents the movement of certain collections of coefficients from each subband in delayed ranked subband grouping 57 (the 'reference' frame) to their new positions so to predict ranked subband grouping 53 (the 'present' frame).
- SBQO- Processing typically proceeds in the order of ranking, that is, from the macro-blocks having the greatest amount of movement to those having the least amount of movement.
- Entirely insignificant subband blocks will not be considered at all and therefore will have no motion vector assigned. This will occur, for example, when there is insignificant or no change at those locations between frames. It can also occur when subbands blocks are zero (0) referenced as described herein above.
- filters used for incremental and fractional movement of one subband block can be used for incremental and fractional movement of another subband block.
- the same advantageous use of separable filters can be carried out in motion prediction block 60.
- motion vectors 58 are output to motion prediction block 60 and positional encoding block 76.
- motion prediction block 60 the motion vectors are used to calculate the shift of certain collections of coefficients from each subband in delayed ranked subband grouping 57 (the 'reference' frame) to their new positions so to predict ranked subband grouping 53 (the 'present' frame).
- the x component of MV2 4 is multiplied by the reciprocal of modulo four (4) and the y component of MV2 4 is multiplied by the reciprocal of modulo two (2).
- FIG. 10 illustrates the movement of all the collections of coefficients to the subband blocks corresponding to SMB2 4. The application of all motion vectors
- MV X x from the collection of motion vectors 58 to delayed ranked subband grouping 57 (the 'reference' frame) results in a prediction of ranked subband grouping 53 (the 'present' frame) and is called the predicted grouping 66 (FIG. 3).
- An alternate embodiment of the above described masking process for determining fractional movement between frames includes the use of 3 x 3 coefficient masks. These masks take a weighted average of the coefficients surrounding a selected coefficient.
- a collection of motion vectors 58 that include only incremental movements is determined as described above and illustrated in FIG. 9(a) and 9(b) for each significant subband block in each subband (SBQO through SB33) or a select number subbands, e.g., SBQO only.
- the collection of motion vectors 58 are fed to motion prediction block 60.
- motion prediction block 60 the collection of motion vectors 58 is applied in a manner analogous to that illustrated in FIG. 10 causing significant subband blocks of the delayed ranked subband grouping 57 to be incrementally shifted. Thereafter, each coefficient of each shifted collection of coefficients has a 3 x 3 mask applied to it. The masks that are applied determine the weighted average of the coefficients surrounding each shifted coefficient. The resultant of that calculation is the prediction of the shifted coefficient, i.e., the coefficients new value.
- motion prediction block 60 After all of the motion vectors from the collection of motion vectors 58 have been applied to delayed ranked subband grouping 57 and all of the coefficients that were shifted by the motion vectors have had the 3 x 3 mask applied to them, the result is output from motion prediction block 60 as predicted grouping 66. Of course, the process is repeated in prediction block 98 of decoder 82 to replicate the masking process carried out in motion prediction block 60.
- predicted grouping 66 is passed to difference block 68 wherein the difference between ranked subband grouping 53 and predicted grouping 66 is determined. As described above, difference block 68 produces grouping differences 70.
- motion compensation methods described herein are illustrated as functioning in conjunction with a tensor product wavelet, it is important to note that the methods can be utilized with other types of transforms. This includes utilizing the motion compensation methods with other transforms in either the time domain or the transform domain.
- data transformed in a DCT can be motion compensated in a manner similar to that described above. That is, the 64 transform coefficients in each of the 8 x 8 blocks of the DCT can be motion compensated in a manner similar to that used to motion compensate the 64 transform coefficients in each of the 8 x 8 subband blocks in SBQO of the tensor product wavelet transform.
- FIG. 11 there is illustrated another embodiment of the video encoder. As in the embodiment described above and illustrated in FIG.
- CIF image 26 is converted to a QCIF image 30 in converter 28.
- QCIF image 30 is transformed and converted to a subband macro-block grouping 40 by the image to feature mapping components 20.
- the collection of subbands 34 and the subband macro-block grouping 40 are converted to ranked subband grouping 53 and ranked subband macro-block grouping 54, respectively, by the components associated with determining the features which changed 22.
- ranked subband macro-block grouping 54 is fed to a motion estimation block 150 and ranked subband grouping 53 is fed to difference block 68.
- an error corrected subband grouping 171 is fed to delay block 156, thereby producing delayed subband grouping 172 (the 'reference frame').
- Such a variation is necessary when quantization (or scaling) is so great that it substantially alters the prediction errors 70 produced in difference block 68.
- a copy of ranked subband grouping 53 is passed unchanged through difference block 68 and stored in memory when the system is referenced to zero (0), for example, when the system is initiated or when the reference in the decoder is to be refreshed. Thereafter, prediction errors 70 are accumulated, i.e., added to the reference, as the prediction errors 70 of each subsequent frame passes quantize block 158.
- the updated reference image is fed to delay block 156, thereby producing delayed subband grouping 172.
- the reference in the encoder remains synchronized with the reference in the decoder.
- motion estimation block 150 and motion prediction block 152 receive the delayed subband grouping 172 from delay block 156, motion estimation and motion prediction are determined by a procedure similar to that described herein above and illustrated in FIGS. 8 through 10.
- a forward feed 159 is provided between change detect 46 and quantize block 158 for adjusting the amount of quantization that is to be performed on a particular block, depending on the amount the block has changed.
- change detect 46 When a large amount of change is detected in change detect 46, a large number of bits are allocated for quantization. And conversely, when a small amount of change is detected in change detect 46, a proportionately lesser number of bits are allocated for quantization.
- FIG. 12 there is illustrated yet another embodiment of the video encoder. The front portion of this embodiment is similar to the embodiments discussed above and illustrated in FIGS. 3 and 11. However, unlike the embodiments described above, motion estimation is carried out in the image domain. This embodiment takes advantage of special hardware configurations presently available on some processors.
- a CIF image 26 is converted to a QCIF image 30 in converter block 28.
- the QCIF image 30 is transformed and converted to a subband macro-block grouping 40 by the image to feature mapping components 20.
- Subband macro-block grouping 40 are processed by the components associated with determining the features which have changed 22 to determine subband macro-block ranking. The results are applied to the collection of subbands 34 resulting in ranked subband grouping 53. Ranked subband grouping 53 is thereafter fed to difference block 68.
- the QCIF image 30, also referred to as the 'present' frame, is also fed to motion estimation block 160 and delay block 166 for determining a collection of motion vectors 162. More specifically, an image frame 30 is delayed in delay 166 producing a delayed image frame 167, also referred to as the 'reference' frame. With reference to FIG. 13, delayed image frame 167 is fed to motion estimation block 160 wherein a P x P pels search region is developed about each significant image macro-block. For example, a P x P pels search region 107 is established about image macro-block 2,4 (IMB2 4). Based on empirical analysis, a search region 107 of 32 x 32 pels is used as the search region about a 16 x 16 pels image macro-block of a QCIF image frame.
- IMB2 4 image macro-block 2,4
- each significant image macro-block (IMB X x ) of the present QCIF image 30 frame is located within the corresponding search region in the delayed image frame 167 for determining the motion vectors.
- IMB2 4 is retrieved from QCIF image 30 and located within search region 107 of delayed image frame 167. This process is analogous to that carried out in the transform domain as described above and illustrated in FIGS. 8 and 9(a).
- IMB2 4 traverses search region 107 seeking to determine the minimum total absolute difference at each step between IMB2 4 and search region 107.
- fractional searching is unnecessary when searching in the image domain. Therefore, after determining the incremental movement of IMB2 4, the x and y coordinates are inversed (multiplied by -1) and stored in memory with the collection of motion vectors 162. The motion vectors are fed to motion prediction block 154 and positional encoding block 76. Thereafter, the motion vectors are applied to delayed subband grouping 172 in a manner similar to that described above with regard to FIGS. 3 and 11 and illustrated in FIG. 10.
- FIG. 14 another embodiment of the video encoder is illustrated wherein the front portion is similar to the embodiments discussed above and illustrated in FIGS. 3, 11 and 12. However, unlike the embodiments described above, both motion estimation and motion prediction are carried out in the image domain.
- a collection of motion vectors 162 are determined in a manner similar to that described above and illustrated in FIGS. 12 and 13. The collection of motion vectors 162 are fed to block 164 for motion prediction and to block 76 for positional encoding.
- an error corrected subband grouping 171 having accumulated errors added thereto, is fed to delay block 156, thereby producing delayed subband grouping 172 (the 'reference frame').
- the delayed subband grouping 172 is then reconstructed by inverse wavelet transform block 174 to form a reconstructed image 176.
- the reconstructed image has a structure that is similar to QCIF image 30 illustrated in FIG. 4(a).
- a portion of the grouping can be reconstructed to effectuate efficiencies. For example, a 3, 5 filter can be used to obtain a reconstructed region having 48 x 48 pels. Regions are selected based on the significance of, i.e., the detected changes within, the image macro-blocks (16 x 16) about which the regions are centered.
- motion prediction block 164 the collection of motion vectors 162 are applied to the reconstructed image 176 (or the reconstructed 48 x 48 pels regions if only regions were inverse wavelet transformed).
- the collection of motion vectors 162 are applied to the reconstructed reference image 176 in a manner analogous to that described above and illustrated in FIG. 10 for the shifting collections of transform coefficients in the subband representation of the QCIF image.
- a prediction 178 is fed to forward wavelet transform block 180 producing predicted grouping 66.
- the predicted grouping 66 is then subtracted from ranked subband grouping 53 in difference block 68 resulting in grouping differences 70.
- Quantization is carried out in block 158 and the errors are accumulated to maintain the reference (as described above) and are also forwarded to positional encoding block 76. Positional encoding of the quantized errors and motion vectors 162 takes place as described above and are forwarded to the decoder via transmission line 80.
- ASIC application-specific integrated circuit
- the principles of the embodiments of the invention could also be implemented in hardware, for example, by means of an application-specific integrated circuit (ASIC).
- the ASIC implementation including the necessary memory requirements, should operate at the pel rate in order to (i) minimize power consumption consistent with the embodiment, and (ii) permit compression of full color video, such as for example a full CCIR601, at a data rate of less than 13.5 MHz.
- the detail terms can be quantized through the use of a bit serial analog-to- digital converter which implements the quantization strategy.
- the resulting bit stream is compressed.
- the photonic/optical device operates, i.e., the number of digital transitions which occur, at the compressed data rate rather than at the image data rate (as in the case of an ASIC) or the processor data rate (as in the case of a conventional processor).
- This will result in an implementation which consumes very little current, thus requiring less power.
- the implementation of an optical method will further reduce power consumption by a factor of approximately ten (10) times that of an ASIC implementation.
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
- Compression, Expansion, Code Conversion, And Decoders (AREA)
- Complex Calculations (AREA)
Applications Claiming Priority (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US6663897P | 1997-11-14 | 1997-11-14 | |
US66638P | 1997-11-14 | ||
PCT/US1998/024189 WO1999026418A1 (en) | 1997-11-14 | 1998-11-13 | Apparatus and method for compressing video information |
Publications (2)
Publication Number | Publication Date |
---|---|
EP1031238A1 EP1031238A1 (en) | 2000-08-30 |
EP1031238A4 true EP1031238A4 (en) | 2003-05-07 |
Family
ID=22070754
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
EP98958556A Withdrawn EP1031238A4 (en) | 1997-11-14 | 1998-11-13 | SYSTEM AND METHOD FOR COMPRESSING VIDEO DATA |
Country Status (7)
Country | Link |
---|---|
EP (1) | EP1031238A4 (ja) |
JP (2) | JP4675477B2 (ja) |
KR (1) | KR100614522B1 (ja) |
CN (1) | CN1190084C (ja) |
AU (1) | AU752219B2 (ja) |
CA (1) | CA2310602C (ja) |
WO (1) | WO1999026418A1 (ja) |
Families Citing this family (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
DE69901525T2 (de) * | 1998-02-13 | 2003-01-09 | Koninklijke Philips Electronics N.V., Eindhoven | Verfahren und vorrichtung zur videokodierung |
FR2813486B1 (fr) * | 2000-08-31 | 2003-01-24 | Canon Kk | Procede et dispositif de transformation de donnees |
KR100366382B1 (ko) * | 2001-02-12 | 2002-12-31 | (주) 멀티비아 | 동영상 부호화 장치 및 방법 |
FI111592B (fi) * | 2001-09-06 | 2003-08-15 | Oulun Yliopisto | Menetelmä ja laite peräkkäisten kuvien koodaamiseksi |
KR100440567B1 (ko) * | 2001-11-06 | 2004-07-21 | 한국전자통신연구원 | 움직임 탐색을 위한 이진 평면 구성방법 및 이를 이용한움직임 추정 장치 |
KR100472476B1 (ko) * | 2002-08-31 | 2005-03-10 | 삼성전자주식회사 | 움직임 보상을 위한 보간 방법 및 그 장치 |
KR100612849B1 (ko) * | 2003-07-18 | 2006-08-14 | 삼성전자주식회사 | 영상 부호화 및 복호화 장치 및 방법 |
JP4556124B2 (ja) * | 2005-02-04 | 2010-10-06 | ソニー株式会社 | 符号化装置および方法、復号装置および方法、画像処理システム、記録媒体、並びにプログラム |
KR100788983B1 (ko) * | 2005-11-02 | 2007-12-27 | 엠텍비젼 주식회사 | 인코딩된 데이터 전달 방법 및 그 방법을 수행하는 촬상장치 |
KR100664550B1 (ko) * | 2005-11-02 | 2007-01-03 | 엠텍비젼 주식회사 | 인코딩된 데이터 전달 방법 및 그 방법을 수행하는 촬상장치 |
KR101277713B1 (ko) * | 2007-02-08 | 2013-06-24 | 삼성전자주식회사 | 비디오 부호화 장치 및 그 방법 |
KR101442278B1 (ko) * | 2007-03-30 | 2014-09-22 | 소니 주식회사 | 정보 처리 장치 및 방법 |
KR101279573B1 (ko) * | 2008-10-31 | 2013-06-27 | 에스케이텔레콤 주식회사 | 움직임 벡터 부호화 방법 및 장치와 그를 이용한 영상 부호화/복호화 방법 및 장치 |
US20120075346A1 (en) * | 2010-09-29 | 2012-03-29 | Microsoft Corporation | Low Complexity Method For Motion Compensation Of DWT Based Systems |
CN106688229B (zh) * | 2014-05-30 | 2019-09-03 | 陈仕东 | 基于变换域传输高清视频的方法 |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5808683A (en) * | 1995-10-26 | 1998-09-15 | Sony Corporation | Subband image coding and decoding |
US5777678A (en) * | 1995-10-26 | 1998-07-07 | Sony Corporation | Predictive sub-band video coding and decoding using motion compensation |
US5764814A (en) * | 1996-03-22 | 1998-06-09 | Microsoft Corporation | Representation and encoding of general arbitrary shapes |
-
1998
- 1998-11-13 CA CA002310602A patent/CA2310602C/en not_active Expired - Fee Related
- 1998-11-13 WO PCT/US1998/024189 patent/WO1999026418A1/en active IP Right Grant
- 1998-11-13 AU AU14577/99A patent/AU752219B2/en not_active Ceased
- 1998-11-13 JP JP2000521650A patent/JP4675477B2/ja not_active Expired - Fee Related
- 1998-11-13 CN CNB988119668A patent/CN1190084C/zh not_active Expired - Fee Related
- 1998-11-13 KR KR1020007005298A patent/KR100614522B1/ko not_active IP Right Cessation
- 1998-11-13 EP EP98958556A patent/EP1031238A4/en not_active Withdrawn
-
2008
- 2008-04-03 JP JP2008096670A patent/JP2008289132A/ja active Pending
Non-Patent Citations (10)
Title |
---|
JIN TAE KIM ET AL: "SUBBAND CODING USING HUMAN VISUAL CHARACTERISTICS FOR IMAGE SIGNALS", IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, IEEE INC. NEW YORK, US, vol. 11, no. 1, 1993, pages 59 - 64, XP000377997, ISSN: 0733-8716 * |
MARTUCCI S A ET AL: "A ZEROTREE WAVELET VIDEO CODER", IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, IEEE INC. NEW YORK, US, vol. 7, no. 1, 1 February 1997 (1997-02-01), pages 109 - 118, XP000678884, ISSN: 1051-8215 * |
NGAN K N ET AL: "VERY LOW BIT RATE VIDEO CODING USING 3D SUBBAND APPROACH", IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, IEEE INC. NEW YORK, US, vol. 4, no. 3, 1 June 1994 (1994-06-01), pages 309 - 316, XP000460762, ISSN: 1051-8215 * |
SCOTTON P ET AL: "A LOW COMPLEXITY VIDEO SUBBAND CODER FOR ATM", SIGNAL PROCESSING. IMAGE COMMUNICATION, ELSEVIER SCIENCE PUBLISHERS, AMSTERDAM, NL, vol. 6, no. 5, 1 October 1994 (1994-10-01), pages 421 - 433, XP000466025, ISSN: 0923-5965 * |
See also references of WO9926418A1 * |
SONG M ET AL: "MOTION ESTIMATION IN DCT DOMAIN", 1996 IEEE INTERNATIONAL CONFERENCE ON COMMUNICATIONS (ICC). CONVERGING TECHNOLOGIES FOR TOMORROW'S APPLICATIONS. DALLAS, JUNE 23 - 27, 1996, IEEE INTERNATIONAL CONFERENCE ON COMMUNICATIONS (ICC), NEW YORK, IEEE, US, vol. 3, 1996, pages 670 - 674, XP000910243, ISBN: 0-7803-3251-2 * |
TAN T K ET AL: "A FREQUENCY SCALABLE CODING SCHEME EMPLOYING PYRAMID AND SUBBAND TECHNIQUES", IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, IEEE INC. NEW YORK, US, vol. 4, no. 2, 1 April 1994 (1994-04-01), pages 203 - 207, XP000489693, ISSN: 1051-8215 * |
UHL A: "Digital image compression based on non-stationary and inhomogeneous multiresolution analyses", PROCEEDINGS OF THE INTERNATIONAL CONFERENCE ON IMAGE PROCESSING (ICIP) AUSTIN, NOV. 13 - 16, 1994, LOS ALAMITOS, IEEE COMP. SOC. PRESS, US, vol. 3 CONF. 1, 13 November 1994 (1994-11-13), pages 378 - 382, XP010146399, ISBN: 0-8186-6952-7 * |
YANG X ET AL: "HIERARCHICAL BACKWARD MOTION COMPENSATION FOR WAVELET VIDEO CODING USING OPTIMIZED INTERPOLATION FILTERS", PROCEEDINGS OF THE INTERNATIONAL CONFERENCE ON IMAGE PROCESSING. ICIP 1997. SANTA BARBARA, CA, OCT. 26 - 29, 1997, LOS ALAMITOS, CA: IEEE, US, vol. 1, 26 October 1997 (1997-10-26), pages 85 - 88, XP000792722, ISBN: 0-8186-8184-5 * |
YOON S H ET AL: "A SCALABLE WAVELET VIDEO CODER FOR HYBRID COMMUNICATION CHANNELS", CONFERENCE RECORD OF THE 31ST ASILOMAR CONFERENCE ON SIGNALS, SYSTEMS & COMPUTERS. PACIFIC GROVE, CA., NOV. 2 - 5, 1997, ASILOMAR CONFERENCE ON SIGNALS, SYSTEMS AND COMPUTERS, LOS ALAMITOS, CA.: IEEE, US, vol. 1, 2 November 1997 (1997-11-02), pages 382 - 386, XP000883963, ISBN: 0-8186-8317-1 * |
Also Published As
Publication number | Publication date |
---|---|
CN1190084C (zh) | 2005-02-16 |
CA2310602C (en) | 2009-05-19 |
JP2001523928A (ja) | 2001-11-27 |
EP1031238A1 (en) | 2000-08-30 |
KR20010032113A (ko) | 2001-04-16 |
JP2008289132A (ja) | 2008-11-27 |
AU1457799A (en) | 1999-06-07 |
CN1281618A (zh) | 2001-01-24 |
CA2310602A1 (en) | 1999-05-27 |
JP4675477B2 (ja) | 2011-04-20 |
KR100614522B1 (ko) | 2006-08-22 |
WO1999026418A1 (en) | 1999-05-27 |
AU752219B2 (en) | 2002-09-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US6937659B1 (en) | Apparatus and method for compressing video information | |
US5272529A (en) | Adaptive hierarchical subband vector quantization encoder | |
CN1115879C (zh) | 利用逐个象素运动估算与帧抽取的图象处理系统 | |
US5786856A (en) | Method for adaptive quantization by multiplication of luminance pixel blocks by a modified, frequency ordered hadamard matrix | |
US5646618A (en) | Decoding one or more variable-length encoded signals using a single table lookup | |
JP2008289132A (ja) | データセットを変換係数に変換する方法 | |
US20120320987A1 (en) | Enhancing fractional pixel motion prediction using a plurality of filters | |
US20050207500A1 (en) | Method and apparatus for motion vector field encoding | |
US20060008000A1 (en) | Fully scalable 3-d overcomplete wavelet video coding using adaptive motion compensated temporal filtering | |
JP2004229315A (ja) | 直交変換−ドメインブロックマッチングを用いる動き推定 | |
JPH11262018A (ja) | 動き補償符号化装置、動き補償符号化方法、及び動き補償符号記録媒体 | |
EP1265443A2 (en) | Wavelet domain motion compensation system | |
WO2005002234A1 (en) | Video coding in an overcomplete wavelet domain | |
CN1078795C (zh) | 用在图象编码系统中的改进的运动补偿方法 | |
JPH09502840A (ja) | テレビ画像サブバンド両立式符号化用エンコーダ/デコーダ及びツリー構造による階層的運動符号化へのその応用 | |
KR20050049517A (ko) | 웨이블릿 기반의 부호화에서 운동 보상된 시간적 필터링을위한 필터링된 영역과 비필터링된 영역 모두를 갖는엘-프레임 | |
Lee et al. | Subband video coding with scene-adaptive hierarchical motion estimation | |
Van Der Auwera et al. | Video coding based on motion estimation in the wavelet detail images | |
KR930006522B1 (ko) | 영상신호의 대역압축 부호화장치 | |
KR0185845B1 (ko) | 웨이브렛 변환기를 이용한 계층 부호화기 | |
MXPA00004526A (en) | Apparatus and method for compressing video information | |
Wang | Fully scalable video coding using redundant-wavelet multihypothesis and motion-compensated temporal filtering | |
Cai | High performance shift invariant motion estimation and compensation in wavelet domain video compression | |
KR20020015231A (ko) | 객체 지향형 영상 부호화 시스템 및 방법 | |
JPH06217291A (ja) | 動き補償フレーム間サブバンド符号化方式 |
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: 20000512 |
|
AK | Designated contracting states |
Kind code of ref document: A1 Designated state(s): BE CH DE DK ES FI FR GB GR IE IT LI NL PT SE |
|
RAP1 | Party data changed (applicant data changed or rights of an application transferred) |
Owner name: ANTEON CORPORATION |
|
RIC1 | Information provided on ipc code assigned before grant |
Ipc: 7H 04N 7/26 B Ipc: 7H 04N 7/32 B Ipc: 7H 04N 7/30 A |
|
A4 | Supplementary search report drawn up and despatched |
Effective date: 20030321 |
|
17Q | First examination report despatched |
Effective date: 20080731 |
|
STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: THE APPLICATION IS DEEMED TO BE WITHDRAWN |
|
18D | Application deemed to be withdrawn |
Effective date: 20130405 |