Pulse or digital communications – Bandwidth reduction or expansion – Television or motion video signal
Reexamination Certificate
1998-11-19
2001-11-20
Kelley, Chris (Department: 2613)
Pulse or digital communications
Bandwidth reduction or expansion
Television or motion video signal
C375S240030, C382S253000
Reexamination Certificate
active
06320907
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates to the field of data processing, and more particularly, to video data processing which stores digital data in dedicated video memories for implementation of real time processing on the digital data, as performed in video decoders using the MPEG compression algorithm or in SQTV/IQTV systems.
BACKGROUND OF THE INVENTION
Reducing memory requirements are important, especially for those systems supporting decoding of the high definition television (HDTV) standard. For example, an MPEG-2 video decoder integrated system capable of decoding high definition sequences, as well as standard definition sequences, normally requires access to an external video memory (DRAM) of at least 80 Mbits through a common interface. Typically, in such an application, the video memory may be configured according to the following frame buffers. A bit buffer for compressed data is set according to the MPEG-2 standard at 9,500,000 bits. An I-frame buffer for the I-picture (Intra-picture) is decompressed in a 4:2:0 format and a P-frame buffer for the decompressed P-picture (Predicted-picture) is in a 4:2:0 format. In addition, a B-frame buffer for the decompressed B-picture (Bidirectionally Predicted Picture) is in a 4:2:0 format. Each frame buffer in the 4:2:0 format may occupy an amount of memory given by:
1920
×
1080
×
8
for
the
luma
Y
=
960
×
540
×
8
for
the
chroma
U
=
960
×
540
×
8
for
the
chroma
V
=
total
Y
+
U
+
V
=
16
,
588
,
800
bit
4
,
147
,
200
bit
4
,
147
,
200
bit
24
,
883
,
200
bit
Therefore, the actual total amount of memory requirement for high definition television decoding will be:
9,500,000+24,883,200+24,883,200+24,883,200=84,149,600 bits
By using fast synchronous memories such as SDRAM, decompression of the B-pictures can be optimized upon reception without storing them. This optimization reduces the external memory requirement to:
9,500,000+24,883,200+24,883,200=59,266,400 bits
In view that the B-buffer is implemented on chip, reduction of the external memory requirements is necessary to convert the scanning of each 8×8 block. This conversion is defined in the MPEG-2 compressed bitstream for each row of the picture (field or frame), as required by the video display processing. Such a conversion macrocell is commonly referred to as a M
ACROBLOCK
To R
ASTER
S
CAN
C
ONVERTER .
Incorporated herein by reference in its entirety is European Patent Application No. 97830041.6, dated Feb. 6, 1997, which is assigned to the assignee of the present invention. This reference discloses an efficient tree-search vector quantization (TSVQ) technique for compressing digital video data to be stored in the external memory. This reduces the video memory requirement of the system. In practice, the memory required by the decoding system can be reduced by recompressing the pictures used as a reference for the prediction (I-, P- and B-pictures) after MPEG decompression, and before storing them in the external video memory. The same TSVQ compression technique is also useful in SQTV processors.
As previously described with respect to the above-referenced patent application, the effectiveness of a compression method based on the TSVQ technique is strongly influenced by the way the quantizer is chosen. The quantizer is used for quantizing the differences among the adjacent pel vectors. The present invention improves the effectiveness of the method described in referenced patent application by providing a more efficient way of selecting the quantizer, apart from other improving features.
As an illustrative example to highlight an object of the present invention, reference is made to the description of the MPEG-2 decoding system disclosed in the referenced patent application. The MPEG-2 decoding system utilizes a tree-search vector quantization technique for compressing digital video data to be stored in an external video memory. In particular, I is a digital picture represented by a matrix of M rows and N columns of pixels. I(y,x) is the pixel corresponding to row y and column x, and is coded with an integer number B of bits (binary digits). The I picture is separated into rectangular blocks having an R×C size (R rows and C columns). The maximum efficiency for the compression is obtained if R and C are chosen among the integer divisors of M and N, respectively.
The algorithm performs a compression of each block by exploiting only the data extracted from the block itself. The compression of each block is a reduction of the number of bits necessary for representing the block itself. This simplifies both access to the block in the stream of compressed data, and also the decompression of the block itself. The TSVQ compression mechanism exploits the correlation existing among adjacent pixels of a picture for reducing the number of bits necessary for a binary description of the picture itself. Considering a vector formed by J rows and K columns, the ideal effectiveness of the compression is obtained if J and K are chosen among the integer dividers of R and C, respectively. It is possible to approximate the value of a pixel vector by appropriately combining only the values of the pixels adjacent to it and forming a second pixel vector. This forms what is commonly referred to as a vector prediction.
It is therefore possible to reduce the amount of binary digits necessary for a digital representation of a picture by defining the prediction mechanism and by appropriately coding only a prediction error vector. This coding of the prediction error vector does not code each pixel. The more precise the vector prediction (and its components), the lower the entropy of the prediction error. That is, the number of bits necessary for coding the prediction error is lower.
In defining a scanning arrangement of the R×C block, for each J×K vector there exists another vector preceding it that may be used as the predictor of the vector itself. An exception is for the first one, which is not subject to any modification with respect to the values of its components. Let
V (i . . . i+J−1, j . . . j+K−1),
i=1, . . . , [R−J+1] and
j=1, . . . , [C−K+1]
be the vector comprising the pixels contained in the rectangular area determined by the following coordinates: top left coordinates (i, j), top right coordinates (i, j+K−1), bottom left coordinates (i+J−1, j), and bottom right coordinates (i+J−1, j+K−1). With respect to the arrangement of
FIG. 9
, the rectangular area is defined as follows:
OV (1 . . . J, 1 . . . K), first scan vector
OV′ (i . . . i+J−1, 1 . . . j+K−1),
i=1, and
j=1, 1+K, 2+K, C−K+1
The prediction error, E=V−V′, is defined according to a scanning arrangement that is divided in regions so that each vector E belongs to only one region. The union of the regions forms the R×C block. A local complexity measure for each region is defined as the average value of the sum of the components of each vector E( ) in terms of its absolute value. Accordingly, j ranges between 1 and (C−K+1)/D, where D is a positive integer ranging between 1 and (C−K+1). Other measures could be used for determining the local complexity measure, such as the maximum value of the components.
A positive value is quantized with one of the G values stored in a table. The positive value represents the centroids of the areas in which the most general scalar statistic of the values have been divided. Such a partition minimizes the mean square error as compared to other possible partitions of the same statistic. Quantization table G is generated by using the Generalized Ll
Bruni Roberta
Pau Danilo
Allen Dyer Doppelt Milbrath & Gilchrist, P.A.
Galanthay Theodore E.
Kelley Chris
Philippe Gims
STMicroelectronics S.r.l.
LandOfFree
Adaptive tree-search vector quantization in MPEG2 decoder does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Adaptive tree-search vector quantization in MPEG2 decoder, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Adaptive tree-search vector quantization in MPEG2 decoder will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2591567