Pulse or digital communications – Bandwidth reduction or expansion – Television or motion video signal
Reexamination Certificate
2001-12-28
2004-07-13
Diep, Nhon (Department: 2613)
Pulse or digital communications
Bandwidth reduction or expansion
Television or motion video signal
Reexamination Certificate
active
06763068
ABSTRACT:
FIELD OF THE INVENTION
The invention relates to a video coding system. In particular, it relates to two standard systems for the compression of video sequences using motion-compensated prediction: ITU-T H.263 and MPEG-4 Very Low Bitrate Video (VLBV).
BACKGROUND OF THE INVENTION
Two standard systems for the compression of video sequences using motion-compensated prediction are: ITU-T H.263, described in
ITU
-
T SG
15
Experts Group for Very Low Bitrate Visual Telephony,
Draft Recommendation H.263, February 1995; and MPEG-4 Very Low Bit rate Video (VLBV), described
MPEG
-4
Video Verification Model—Version
5.0, by the MPEG Video Group, Doc. ISO/IEC/JTCI/SC29/WG11, N1469 Maceio, November 1996. There have been some extensions to H.263 after the definition of Version 1. The extended version is often referred to as H.263+. The term H.263 is used here to refer to the un-extended version, i.e. Version 1. Because of the similarity between the algorithms employed in MPEG-4 VLBV and in H.263, the discussion here will focus on algorithms for H.263.
H.263 is an ITU-T recommendation for the compression of video sequences at low bit rates (<64 kbits/sec), and is based on an earlier ITU-T recommendation, H.261. A block diagram of the control components of an H.263 encoder is depicted in FIG.
1
. (The video coder is not shown.) The main elements of such an encoder are a prediction module
11
, a pair of block transformation modules
12
(the transform modules T and T
−
), and a pair of quantization modules
13
(the quantizer modules Q and Q
−
). In addition, there is a coding control module
14
.
The coding control module
14
determines all coding parameters; it acts as the brain of the system. The INTRA/INTER decision data flow from the coding control signals whether the knowledge of previous frames will be exploited for encoding the current frame.
The quantization indication data flow provided by the coding control module determines what are called quantizer parameters to be used for each macroblock. The determination can be at the frame level or at the macroblock level. The invention focuses on the generation of this signal.
The ‘video multiplex coder’ referred to in
FIG. 1
is simply a multiplexer, and does not use ‘video in’ as one of its inputs. The signals indicated as being provided “to video multiplex coder” comprise the compressed representation of a video signal.
To exploit the temporal correlation between successive frames, the system first performs a motion-compensated prediction if previously reconstructed frames are available and useful, the term ‘useful’ indicating that using the previous frames (INTER coding) would yield better compression performance than not using them (INTRA coding). If the consequent frames are not sufficiently correlated, it may be possible that INTRA coding yields better compression performance. The system first performs the motion-compensated prediction using motion information and the previously reconstructed frames. The motion information (v in
FIG. 1
) is transmitted (from encoder to decoder) in order to allow the decoder to perform the same prediction as the prediction performed in the encoder. Next, a block transform, referred to as a DCT (i.e. a Discrete Cosine Transform), is applied to the prediction error (or the frame itself in the case of no prediction) in the block T to exploit the spatial correlation. Finally, the DCT coefficients of the prediction error are quantized and entropy coded in the block Q. The quantizer is the main mechanism for introducing loss to the video sequence to achieve higher compression. The amount of loss (and thus the bit rate) is controlled by the quantizer step size, which in turn is parameterized by a quantizer parameter (QP), having integer values between 1 and 31 and provided by the coding control module
14
to the block Q. QP must be known to the decoder, so it is transmitted as side (ancillary) information (and designated as qz in FIG.
1
).
A fundamental layer of H.263 is the macroblock layer. A macroblock (MB) is the basic building block of H.263 in the sense that the main elements of encoding (prediction, block transformation and quantization) can be performed by processing one macroblock at a time. A macroblock consists of the representations of a 16×16 luminance block, two 8×8 chrominance blocks, and macroblock-level coding parameters (such as macroblock type, etc.); some macroblock-level coding parameters are optional. Macroblocks are transmitted in raster scan order. The macroblock-level coding parameters are encoded at the macroblock level only when there is a need to do so, since they are costly in terms of bits. (There is a frame-level encoding of the coding parameters such as the quantization parameter QP. When macroblock-level encoding is not performed, these frame level values are used.)
H.263 provides limited macroblock-level control of QP; there is an optional 2-bit DQUANT field that encodes the difference between the QP of the current macroblock and the QP of the previously encoded macroblock. (See Section 5.3.6 of H.263+(February 1998).) Due to the bit-field restriction described in section 5.3.6 of H.263, the macroblock QP can be varied by at most ±2 each time it is changed. In a scenario where QP variation is used for rate control, such a restriction on the range of variation of QP is quite sufficient. However, there may be other reasons to vary QP besides rate control, such as region-of-interest coding, which is a technique of allocating more bits (and thus introducing less loss) to an automatically-detected or user-defined region of interest, such as a human face.
The Problem Solved by the Invention
Consider a scenario for which macroblock-level QP variation is needed for a purpose other than rate control, such as region-of-interest (ROI) coding. In such a scenario, limited macroblock-level control of QP poses a significant restriction. An arbitrary QP distribution either suggested by a region-of-interest analyzer or input by a user cannot be fully realized by an H.263 (or MPEG-4 VLBV) encoder. Thus, such an encoder needs to choose an approximate realization that is as close to the originally suggested distribution as possible, in some defined sense. Given a definition of optimality (or equivalently a measure of cost), the invention provides a method to optimally choose the realization, i.e. it provides a method to minimize the total cost incurred by constrained realization, the total cost being defined so as to be lower the closer the QP distribution is to the suggested distribution, but higher the more bits that must be used.
As a related problem, in H.263+, defined in
ITU
-
T SG
15
Experts Group for Very Low Bitrate Visual Telephony,
Draft Recommendation H.263 Version 2, January 1998, an exact representation of an arbitrary distribution may be very costly in terms of a bit budget, and hence may not be the most desirable solution. Quantization can be varied on a macroblock basis, and no finer variation is possible. By an exact representation of an arbitrary distribution is meant an arbitrary selection of macroblock QPs. Consider a vector QP, formed by assigning a separate QP for each macroblock. Representing some of the component vectors cost much less (in bits) than representing others. If a QP is chosen for each macroblock without considering the QP values of the neighboring macroblocks, it is likely that a prohibitively high number of bits will be spent in representing the variation of QP from block to block.
An encoder needs to find the best trade-off between following the original suggestion (for a QP distribution) by a region of interest analyzer on the one hand, and minimizing the total bits spent for QP variation on the other. For this related problem, the invention provides a method for finding the best trade-off when the cost function is defined to represent the trade-off.
How the Problem was Solved Earlier
According to the prior art, for MPEG-4 video encoders employing ROI coding, QP control is achieved through th
Diep Nhon
Ware Fressola Van Der Sluys & Adolphson
LandOfFree
Method and apparatus for selecting macroblock quantization... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method and apparatus for selecting macroblock quantization..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and apparatus for selecting macroblock quantization... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3207825