Patent 08090023 - Method and apparatus for variable accuracy inter-picture timing specification for digital video encoding with reduced requirements for division operations > Description
This application is a continuation of U.S. patent application Ser. No. 10/792,514, entitled “METHOD AND APPARATUS FOR VARIABLE ACCURACY INTER-PICTURE TIMING SPECIFICATION FOR DIGITAL VIDEO ENCODING WITH REDUCED REQUIREMENTS FOR DIVISION OPERATIONS”, filed Mar. 2, 2004, now issued as U.S. Pat. No. 7,339,991. U.S. patent application Ser. No. 10/792,514 is a continuation application of U.S. patent application Ser. No. 10/313,773, now issued as U.S. Pat. No. 6,728,315, entitled “METHOD AND APPARATUS FOR VARIABLE ACCURACY INTER-PICTURE TIMING SPECIFICATION FOR DIGITAL VIDEO ENCODING WITH REDUCED REQUIREMENTS FOR DIVISION OPERATIONS”, filed Dec. 6, 2002. U.S. patent application Ser. No. 10/313,773 claims priority to U.S. Provisional Application 60/398,625, filed Jul. 24, 2002. U.S. patent application Ser. No. 10/792,514, now issued as U.S. Pat. No. 7,339,991, U.S. patent application Ser. No. 10/313,773, now issued as U.S. Pat. No. 6,728,315, and U.S. Provisional Patent Application 60/398,625 are incorporated herein by reference.
This Application is related to U.S. patent application 10/313,773, filed Dec. 6, 2002, now issued as U.S. Pat. No. 6,728,315; U.S. patent application 10/792,514, filed Mar. 2, 2004, now issued as U.S. Pat. No. 7,339,991; U.S. patent application 11/621,977, filed Jan. 10, 2007, now published as U.S. Publication 2007/0183501; U.S. patent application 11/621,979, filed Jan. 10, 2007, now published as U.S. Publication 2007/0183502; U.S. patent application 10/291,320, filed Nov. 8, 2002, now issued as U.S. Pat. No. 7,088,776; U.S. patent application 10/792,669, filed Mar. 2, 2004, now issued as U.S. Pat. No. 7,292,636; U.S. patent application 11/621,969, filed Jan. 10, 2007, now issued as U.S. Pat. No. 7,903,730; U.S. patent application 11/621,971, filed Jan. 10, 2007, now issued as U.S. Pat. No. 7,826,528; and U.S. patent application 11/621,974, filed Jan. 10, 2007, now issued as U.S. Pat. No. 7,668,240.
The present invention relates to the field of multimedia compression systems. In particular the present invention discloses methods and systems for specifying variable accuracy inter-picture timing with reduced requirements for processor intensive division operation.
Digital based electronic media formats are finally on the cusp of largely replacing analog electronic media formats. Digital compact discs (CDs) replaced analog vinyl records long ago. Analog magnetic cassette tapes are becoming increasingly rare. Second and third generation digital audio systems such as Mini-discs and MP3 (MPEG Audio—layer 3) are now taking market share from the first generation digital audio format of compact discs.
The video media formats have been slower to move to digital storage and digital transmission formats than audio media. The reason for this slower digital adoption has been largely due to the massive amounts of digital information required to accurately represent acceptable quality video in digital form and the fast processing capabilities needed to encode compressed video. The massive amounts of digital information needed to accurately represent video require very high-capacity digital storage systems and high-bandwidth transmission systems.
However, video is now rapidly moving to digital storage and transmission formats. Faster computer processors, high-density storage systems, and new efficient compression and encoding algorithms have finally made digital video transmission and storage practical at consumer price points. The DVD (Digital Versatile Disc), a digital video system, has been one of the fastest selling consumer electronic products in years. DVDs have been rapidly supplanting Video-Cassette Recorders (VCRs) as the pre-recorded video playback system of choice due to their high video quality, very high audio quality, convenience, and extra features. The antiquated analog NTSC (National Television Standards Committee) video transmission system is currently in the process of being replaced with the digital ATSC (Advanced Television Standards Committee) video transmission system.
Computer systems have been using various different digital video encoding formats for a number of years. Specifically, computer systems have employed different video coder/decoder methods for compressing and encoding or decompressing and decoding digital video, respectively. A video coder/decoder method, in hardware or software implementation, is commonly referred to as a “CODEC”.
Among the best digital video compression and encoding systems used by computer systems have been the digital video systems backed by the Motion Pictures Expert Group commonly known by the acronym MPEG. The three most well known and highly used digital video formats from MPEG are known simply as MPEG-1, MPEG-2, and MPEG-4. VideoCDs (VCDs) and early consumer-grade digital video editing systems use the early MPEG-1 digital video encoding format. Digital Versatile Discs (DVDs) and the Dish Network brand Direct Broadcast Satellite (DBS) television broadcast system use the higher quality MPEG-2 digital video compression and encoding system. The MPEG-4 encoding system is rapidly being adapted by the latest computer based digital video encoders and associated digital video players.
The MPEG-2 and MPEG-4 standards compress a series of video frames or video fields and then encode the compressed frames or fields into a digital bitstream. When encoding a video frame or field with the MPEG-2 and MPEG-4 systems, the video frame or field is divided into a rectangular grid of pixelblocks. Each pixelblock is independently compressed and encoded.
When compressing a video frame or field, the MPEG-4 standard may compress the frame or field into one of three types of compressed frames or fields: Intra-frames (I-frames), Unidirectional Predicted frames (P-frames), or Bi-Directional Predicted frames (B-frames). Intra-frames completely independently encode an independent video frame with no reference to other video frames. P-frames define a video frame with reference to a single previously displayed video frame. B-frames define a video frame with reference to both a video frame displayed before the current frame and a video frame to be displayed after the current frame. Due to their efficient usage of redundant video information, P-frames and B-frames generally provide the best compression.
A method and apparatus for performing motion estimation in a video codes is disclosed. Specifically, the present invention discloses a system that quickly calculates estimated motion vectors in a very efficient manner without requiring an excessive number of division operations.
In one embodiment, a first multiplicand is determined by multiplying a first display time difference between a first video picture and a second video picture by a power of two scale value. This step scales up a numerator for a ratio. Next, the system determines a scaled ratio by dividing that scaled numerator by a second first display time difference between said second video picture and a third video picture. The scaled ratio is then stored to be used later for calculating motion vector estimations. By storing the scaled ratio, all the estimated motion vectors can be calculated quickly with good precision since the scaled ratio saves significant bits and reducing the scale is performed by simple shifts thus eliminating the need for time consuming division operations.
Other objects, features, and advantages of present invention will be apparent from the company drawings and from the following detailed description.
The objects, features, and advantages of the present invention will be apparent to one skilled in the art, in view of the following detailed description in which:
A method and system for specifying Variable Accuracy Inter-Picture Timing in a multimedia compression and encoding system with reduced requirements for division operations is disclosed. In the following description, for purposes of explanation, specific nomenclature is set forth to provide a thorough understanding of the present invention. However, it will be apparent to one skilled in the art that these specific details are not required in order to practice the present invention. For example, the present invention has been described with reference to the MPEG multimedia compression and encoding system. However, the same techniques can easily be applied to other types of compression and encoding systems.
A Discrete Cosine Transformation (DCT) unit 110 processes each pixelblock in the video frame. The frame may be processed independently (an intra-frame) or with reference to information from other frames received from the motion compensation unit (an inter-frame). Next, a Quantizer (Q) unit 120 quantizes the information from the Discrete Cosine Transformation unit 110. Finally, the quantized video frame is then encoded with an entropy encoder (H) unit 180 to produce an encoded bitstream. The entropy encoder (H) unit 180 may use a variable length coding (VLC) system.
Since an inter-frame encoded video frame is defined with reference to other nearby video frames, the digital video encoder 100 needs to create a copy of how each decoded frame will appear within a digital video decoder such that inter-frames may be encoded. Thus, the lower portion of the digital video encoder 100 is actually a digital video decoder system. Specifically, an inverse quantizer (Q−1) unit 130 reverses the quantization of the video frame information and an inverse Discrete Cosine Transformation (DCT−1) unit 140 reverses the Discrete Cosine Transformation of the video frame information. After all the DCT coefficients are reconstructed from inverse Discrete Cosine Transformation (DCT−1) unit 140, the motion compensation unit will use that information, along with the motion vectors, to reconstruct the encoded video frame. The reconstructed video frame is then used as the reference frame for the motion estimation of the later frames.
The decoded video frame may then be used to encode inter-frames (P-frames or B-frames) that are defined relative to information in the decoded video frame. Specifically, a motion compensation (MC) unit 150 and a motion estimation (ME) unit 160 are used to determine motion vectors and generate differential values used to encode inter-frames.
A rate controller 190 receives information from many different components in a digital video encoder 100 and uses the information to allocate a bit budget for each video frame. The rate controller 190 should allocate the bit budget in a manner that will generate the highest quality digital video bit stream that that complies with a specified set of restrictions. Specifically, the rate controller 190 attempts to generate the highest quality compressed video stream without overflowing buffers (exceeding the amount of available memory in a video decoder by sending more information than can be stored) or underflowing buffers (not sending video frames fast enough such that a video decoder runs out of video frames to display).
In some video signals the time between successive video pictures (frames or fields) may not be constant. (Note: This document will use the term video pictures to generically refer to video frames or video fields.) For example, some video pictures may be dropped because of transmission bandwidth constraints. Furthermore, the video timing may also vary due to camera irregularity or special effects such as slow motion or fast motion. In some video streams, the original video source may simply have non-uniform inter-picture times by design. For example, synthesized video such as computer graphic animations may have non-uniform timing since no arbitrary video timing is imposed by a uniform timing video capture system such as a video camera system. A flexible digital video encoding system should be able to handle non-uniform video picture timing.
As previously set forth, most digital video encoding systems partition video pictures into a rectangular grid of pixelblocks. Each individual pixelblock in a video picture is independently compressed and encoded. Some video coding standards, e.g., ISO MPEG or ITU H.264, use different types of predicted pixelblocks to encode video pictures. In one scenario, a pixelblock may be one of three types:
- 1. I-pixelblock—An Intra (I) pixelblock uses no information from any other video pictures in its coding (it is completely self-defined);
- 2. P-pixelblock—A unidirectionally predicted (P) pixelblock refers to picture information from one preceding video picture; or
- 3. B-pixelblock—A bi-directional predicted (B) pixelblock uses information from one preceding picture and one future video picture.
If all the pixelblocks in a video picture are Intra-pixelblocks, then the video picture is an Intra-frame. If a video picture only includes unidirectional predicted macro blocks or intra-pixelblocks, then the video picture is known as a P-frame. If the video picture contains any bi-directional predicted pixelblocks, then the video picture is known as a B-frame. For the simplicity, this document will consider the case where all pixelblocks within a given picture are of the same type.
An example sequence of video pictures to be encoded might be represented as:
I1 B2 B3 B4 P5 B6 B7 B8 B9 P10 B11 P12 B13 I14 . . .
where the letter (I, P, or B) represents if the video picture is an I-frame, P-frame, or B-frame and the number represents the camera order of the video picture in the sequence of video pictures. The camera order is the order in which a camera recorded the video pictures and thus is also the order in which the video pictures should be displayed (the display order).
The previous example series of video pictures is graphically illustrated in
In the scenario of
Since B-pictures use information from future pictures (pictures that will be displayed later), the transmission order is usually different than the display order. Specifically, video pictures that are needed to construct other video pictures should be transmitted first. For the above sequence, the transmission order might be:
I1 P5 B2 B3 B4 P10 B6 B7 B8 B9 P12 B11 I14 B13 . . .
The coding of the P-frame pictures typically utilizes Motion Compensation, wherein a Motion Vector is computed for each pixelblock in the picture. Using the computed motion vector, a prediction pixelblock (P-pixelblock) can be formed by translation of pixels in the aforementioned previous picture. The difference between the actual pixelblock in the P-frame picture and the prediction pixelblock is then coded for transmission.
The coding of P-Pictures typically utilize Motion Compensation (MC), wherein a Motion Vector (MV) pointing to a location in a previous picture is computed for each pixelblock in the current picture. Using the motion vector, a prediction pixelblock can be formed by translation of pixels in the aforementioned previous picture. The difference between the actual pixelblock in the P-Picture and the prediction pixelblock is then coded for transmission.
Each motion vector may also be transmitted via predictive coding. For example, a motion vector prediction may be formed using nearby motion vectors. In such a case, then the difference between the actual motion vector and the motion vector prediction is coded for transmission.
Each B-pixelblock uses two motion vectors: a first motion vector referencing the aforementioned previous video picture and a second motion vector referencing the future video picture. From these two motion vectors, two prediction pixelblocks are computed. The two predicted pixelblocks are then combined together, using some function, to form a final predicted pixelblock. As above, the difference between the actual pixelblock in the B-frame picture and the final predicted pixelblock is then encoded for transmission.
As with P-pixelblocks, each motion vector (MV) of a B-pixelblock may be transmitted via predictive coding. Specifically, a predicted motion vector is formed using nearby motion vectors. Then, the difference between the actual motion vector and the predicted is coded for transmission.
However, with B-pixelblocks the opportunity exists for interpolating motion vectors from motion vectors in the nearest stored picture pixelblock. Such motion vector interpolation is carried out both in the digital video encoder and the digital video decoder.
This motion vector interpolation works particularly well on video pictures from a video sequence where a camera is slowly panning across a stationary background. In fact, such motion vector interpolation may be good enough to be used alone. Specifically, this means that no differential information needs be calculated or transmitted for these B-pixelblock motion vectors encoded using interpolation.
To illustrate further, in the above scenario let us represent the inter-picture display time between pictures i and j as Di,j, i.e., if the display times of the pictures are Ti and Tj, respectively, then
Di,j=Ti−Tj from which it follows that
Note that Di,j may be negative in some cases.
Thus, if MV5,1 is a motion vector for a P5 pixelblock as referenced to I1, then for the corresponding pixelblocks in B2, B3 and B4 the motion vectors as referenced to I1 and P5, respectively would be interpolated by
Note that since ratios of display times are used for motion vector prediction, absolute display times are not needed. Thus, relative display times may be used for Di,j inter-picture display time values.
This scenario may be generalized, as for example in the H.264 standard. In the generalization, a P or B picture may use any previously transmitted picture for its motion vector prediction. Thus, in the above case picture B3 may use picture I1 and picture B2 in its prediction. Moreover, motion vectors may be extrapolated, not just interpolated. Thus, in this case we would have:
Such motion vector extrapolation (or interpolation) may also be used in the prediction process for predictive coding of motion vectors.
The variable inter-picture display times of video sequences should be encoded and transmitted in a manner that renders it possible to obtain a very high coding efficiency and has selectable accuracy such that it meets the requirements of a video decoder. Ideally, the encoding system should simplify the tasks for the decoder such that relatively simple computer systems can decode the digital video.
The variable inter-picture display times are potentially needed in a number of different video encoding systems in order to compute differential motion vectors, Direct Mode motion vectors, and/or Implicit B Prediction Block Weighting.
The problem of variable inter-picture display times in video sequences is intertwined with the use of temporal references. Ideally, the derivation of correct pixel values in the output pictures in a video CODEC should be independent of the time at which that picture is decoded or displayed. Hence, timing issues and time references should be resolved outside the CODEC layer.
There are both coding-related and systems-related reasons underlying the desired time independence. In a video CODEC, time references are used for two purposes:
(1) To establish an ordering for reference picture selection; and
(2) To interpolate motion vectors between pictures.
To establish an ordering for reference picture selection, one may simply send a relative position value. For example, the difference between the frame position N in decode order and the frame position M in the display order, i.e., N−M. In such an embodiment, time-stamps or other time references would not be required. To interpolate motion vectors, temporal distances would be useful if the temporal distances could be related to the interpolation distance. However, this may not be true if the motion is non-linear. Therefore, sending parameters other than temporal information for motion vector interpolation seems more appropriate.
In terms of systems, one can expect that a typical video CODEC is part of a larger system where the video CODEC coexists with other video (and audio) CODECs. In such multi-CODEC systems, good system layering and design requires that general functions, which are logically CODEC-independent such as timing, be handled by the layer outside the CODEC. The management of timing by the system and not by each CODEC independently is critical to achieving consistent handling of common functions such as synchronization. For instance in systems that handle more than one stream simultaneously, such as a video/audio presentation, timing adjustments may sometimes be needed within the streams in order to keep the different streams synchronized. Similarly, in a system that handles a stream from a remote system with a different clock timing adjustments may be needed to keep synchronization with the remote system. Such timing adjustments may be achieved using time stamps. For example, time stamps that are linked by means of “Sender Reports” from the transmitter and supplied in RTP in the RTP layer for each stream may be used for synchronization. These sender reports may take the form of:
Video RTP TimeStamp X is aligned with reference timestamp Y
Audio RTP TimeStamp W is aligned with reference timestamp Z
Wherein the wall-clock rate of the reference timestamps is known, allowing the two streams to be aligned. However, these timestamp references arrive both periodically and separately for the two streams, and they may cause some needed re-alignment of the two streams. This is generally achieved by adjusting the video stream to match the audio or vice-versa. System handling of time stamps should not affect the values of the pixels being displayed. More generally, system handling of temporal information should be performed outside the CODEC.
A Specific Example
As set forth in the previous section, the problem in the case of non uniform inter-picture times is to transmit the inter-picture display time values Di,j to the digital video receiver in an efficient manner. One method of accomplishing this goal is to have the system transmit the display time difference between the current picture and the most recently transmitted stored picture for each picture after the first picture. For error resilience, the transmission could be repeated several times within the picture. For example, the display time difference may be repeated in the slice headers of the MPEG or H.264 standards. If all slice headers are lost, then presumably other pictures that rely on the lost picture for decoding information cannot be decoded either.
Thus, with reference to the example of the preceding section, a system would transmit the following inter-picture display time values:
D5,1 D2,5 D3,5 D4,5 D10,5 D6,10 D7,10 D8,10 D9,10 D12,10 D11,12 D14,12 D13,14 . . .
For the purpose of motion vector estimation, the accuracy requirements for the inter-picture display times Di,j may vary from picture to picture. For example, if there is only a single B-frame picture B6 halfway between two P-frame pictures P5 and P7, then it suffices to send only:
D7,5=2 and D6,7=−1
where the Di,j inter-picture display time values are relative time values.
If, instead, video picture B6 is only one quarter the distance between video picture P5 and video picture P7 then the appropriate Di,j inter-picture display time values to send would be:
D7,5=4 and D6,7=−1
Note that in both of the preceding examples, the display time between the video picture B6 and video picture video picture P7 (inter-picture display time D6,7) is being used as the display time “unit” value. In the most recent example, the display time difference between video picture P5 and picture video picture P7 (inter-picture display time D6,7) is four display time “units” (4*D6,7)
Improving Decoding Efficiency
In general, motion vector estimation calculations are greatly simplified if divisors are powers of two. This is easily achieved in our embodiment if Di,j (the inter-picture time) between two stored pictures is chosen to be a power of two as graphically illustrated in
In the case where an inter-picture time is to be a power of two, the number of data bits can be reduced if only the integer power (of two) is transmitted instead of the full value of the inter-picture time.
Alternatively, the motion vector interpolation of extrapolation operation can be approximated to any desired accuracy by scaling in such a way that the denominator is a power of two. (With a power of two in the denominator division may be performed by simply shifting the bits in the value to be divided.) For example,
Where the value P is a power of two and Z5,4=P*D5,4/D5,1 is rounded or truncated to the nearest integer. The value of P may be periodically transmitted or set as a constant for the system. In one embodiment, the value of P is set as P=28=256.
The advantage of this approach is that the decoder only needs to compute Z5,4 once per picture or in many cases the decoder may pre-compute and store the Z value. This allows the decoder to avoid having to divide by D5,1 for every motion vector in the picture such that motion vector interpolation may be done much more efficiently. For example, the normal motion vector calculation would be:
But if we calculate and store Z5,4 wherein Z5,4=P*D5,4/D5,1 then
But since the P value has been chosen to be a power of two, the division by P is merely a simple shift of the bits. Thus, only a single multiplication and a single shift are required to calculate motion vectors for subsequent pixelblocks once the Z value has been calculated for the video picture. Furthermore, the system may keep the accuracy high by performing all divisions last such that significant bits are not lost during the calculation. In this manner, the decoder may perform exactly the same as the motion vector interpolation as the encoder thus avoiding any mismatch problems that might otherwise arise.
Since division (except for division by powers of two) is a much more computationally intensive task for a digital computer system than addition or multiplication, this approach can greatly reduce the computations required to reconstruct pictures that use motion vector interpolation or extrapolation.
In some cases, motion vector interpolation may not be used. However, it is still necessary to transmit the display order of the video pictures to the receiver/player system such that the receiver/player system will display the video pictures in the proper order. In this case, simple signed integer values for Di,j suffice irrespective of the actual display times. In some applications only the sign (positive or negative) may be needed to reconstruct the picture ordering.
The inter-picture times Di,j may simply be transmitted as simple signed integer values. However, many methods may be used for encoding the Di,j values to achieve additional compression. For example, a sign bit followed by a variable length coded magnitude is relatively easy to implement and provides coding efficiency.
One such variable length coding system that may be used is known as UVLC (Universal Variable Length Code). The UVLC variable length coding system is given by the code words:
Another method of encoding the inter-picture times may be to use arithmetic coding. Typically, arithmetic coding utilizes conditional probabilities to effect a very high compression of the data bits.
Thus, the present invention introduces a simple but powerful method of encoding and transmitting inter-picture display times and methods for decoding those inter-picture display times for use in motion vector estimation. The encoding of inter-picture display times can be made very efficient by using variable length coding or arithmetic coding. Furthermore, a desired accuracy can be chosen to meet the needs of the video codec, but no more.
The foregoing has described a system for specifying variable accuracy inter-picture timing in a multimedia compression and encoding system. It is contemplated that changes and modifications may be made by one of ordinary skill in the art, to the materials and arrangements of elements of the present invention without departing from the scope of the invention.