Pulse or digital communications – Bandwidth reduction or expansion – Television or motion video signal
Reexamination Certificate
2000-12-21
2003-09-30
Le, Vu (Department: 2613)
Pulse or digital communications
Bandwidth reduction or expansion
Television or motion video signal
C375S240240, C348S425300
Reexamination Certificate
active
06628709
ABSTRACT:
DESCRIPTION OF RELATED ART
1. Field of the Invention
This invention relates to video formatting and the like. More specifically the invention relates to efficient conversion between transmission or storage formats and editing formats. The invention is particularly focused on conversion between DV and MPEG formatting.
2. Background of the Invention
DV is an SMPTE (Society of Motion Picture and Television Engineers) standardized digital video format targeted at acquisition and editing of professional quality video. A number of equipment vendors have adopted DV in their digital video products. Because MPEG is another popular digital video format standardized by ISO (International Standardization Organization), the industry has been making an effort to promote the interoperability between DV and MPEG by developing a software-only real-time transcoder between those two formats. A fundamental problem with transcoding between the DV and MPEG formats is bit number prediction. This typically arises in I frame-based MPEG2 to DV transcoder development as well as in DV encoding process generally.
In both DV and MPEG, video image data is compressed by transforming it into the frequency domain, using a discrete cosine transform (DCT), and by variable length coding (VLC) the transformed data. To best serve its target applications, DV is a DCT/VLC-based, I-frame only fixed frame-length format. To best serve tape recording replaying and to provide synchronized transmission, each frame is further broken down into many “segments”, each consisting of five “macroblocks”. The DV macroblocks are similar to MPEG macroblocks, with several exceptions, including that DV each segment has a fixed bit length while there is no such constraint in MPEG. Because adjacent macroblocks have strong correlation and usually have similar coding complexity, the positions of the five macroblocks in each segment are sparsely scattered in the frame according to a pseudo-random deterministic pattern. This reduces the possibility of “bit overflow” or “bit vacancy,” Which would likely occur if the five macroblocks of a segment were close to each other and had similar complexity. This patterning of the macroblocks would produce either a large number of bits or very few bits in each segment.
Because each DV segment has a fixed bit budget, if the coding process produces more bits than a segment can accommodate, some high frequency DCT coefficients will be discarded. To control the number of bits produced, each macroblock is assigned a quantization number. Quantization numbers fall in the range of 0-15, with 15 corresponding to the finest quantizer and 0 to the coarsest. The quantization number is the index of the pre-determined 8×8 “quantization matrices.” These matrices consist of the quantization steps for each DCT coefficient.
A fundamental problem faced in the encoding/transcoding process is how to achieve the best picture quality given the fixed bit space assigned to each segment. Naturally it would be preferable to fully utilize this bit budget, as any unused bit space can not be used by other segments and, thus, is wasted. It is not proper, however, to use always the finest quantizer and produce the greatest number of bits, even though with this strategy there will not be any wasted bit space. If all the VLC bits can fit in the budget with the finest quantizer (quantization number=15) then that is the optimal solution; otherwise, there are two directions to go to within the range of all possible quantizers. The two directions are either (1) to choose a finer quantizer and discard high frequency bits that can not fit in, or (2) to choose a coarser quantizer to fit in high frequency bits at the expense of increasing the quantization error of the lower frequency coefficients.
Currently the solution to the above problem that has been adopted by the majority of DV encoders in the industry is to select the quantization numbers, which just fit in all the bits into the given space. Any finer quantizer will cause the coefficients to be discarded. Experiments prove that this scheme produces better picture quality than simply using the finest quantizer at all times, although it can be shown that this solution is still not optimal.
At present the criterion for selecting a quantization number is based on the fixed space criterion. The non-optimality of the prior art criteria for quantization table selection is the result of the use of a fixed space for all bits. The selected quantization number N should be such that, with quantization number N all the bits can fit in the fixed space, but with quantization of N+1 (if N is not the maximum possible number 15, which corresponds to the finest quantizer), there will be some overflow bits. This criterion is referred to as the “best fit” criterion.
The best fit criterion does not necessarily produce the best picture quality given the fixed bit space. Without the loss of generality, suppose that just one DCT block is to be coded, given the bit space according to the “best fit” criterion. In this example, all of the AC coefficients for the DCT can be written in a line according to their scan order. In this representation the first line is the coefficient index, the second line is the coefficient values and the third line is the area number.
TABLE 1
1
2
3
. . .
8
. . .
23
. . .
45
. . .
62
63
C1
C2
C3
. . .
C8
. . .
C23
. . .
C45
. . .
C62
C63
0
1
2
3
Suppose we have the following specific DCT block with coefficient values of 1 for coefficients 45 and 62.
TABLE 2
1
2
3
. . .
8
. . .
23
. . .
45
. . .
62
63
C1
C2
C3
. . .
C8
. . .
C23
. . .
1
0
1
0
0
1
2
3
If with quantization number QN equals M, the quantization step for area 3 is 1 and a coefficient 62 must be discarded because of the limited space. But if QN is M−1, the quantization step for the area becomes 2 (the quantization steps for other areas are unchanged) and all the bits can fit in the space allocated. According to the “best fit” criterion, therefore, QN should be M−1. Because of the larger step size when QN is M−1, both coefficient 45 and 62 become zero and are effectively discarded. When the quantization number M chosen, however, coefficient 45 in still encoded although coefficient 62 is not. Clearly QN equal to M does better than M−1 and “best fit” criterion does not provide the best possible result for this block.
The typical method used to find the “best fit” quantizer, is to try all the candidates one by one, beginning with the finest quantizer, until all of the quantized bits fit into the allotted space. Each try requires a quantization, a run-length scan and many VLC table look-ups to determine the number of bits that are going to be produced from the VLC stage. Given that each frame has hundreds of segments and assuming a frame rate of 30 frames/second, this brute force approach becomes a bottleneck for a real-time software implementation, or a bulky part in a hardware implementation of the encoder.
The DV standard and consists of two sub-standards: DV25 and DV50. DV25 is a 4:1:1 format for semi-professional quality video while DV50 is a 4:2:2 format for professional, studio quality video. The industry has also developed DVHD format for the HDTV applications, based on the DV standard. The bit rates for this standard are 25 Mbps, 50 Mbps and 100 Mbps (DVHD) respectively. These formats are very similar to each other and can be understood through a description of DV50.
As in many video standards, in DV50 each frame is divided into 8×8 DCT blocks and every four DCT blocks (2 luminance 8×8 DCT blocks and two chrominance 8×8 DCT blocks) constitute a macroblock in the pattern illustrated in FIG.
1
.
There are two DCT modes in DV; 88-DCT and 248-DCT. 88-DCT is just the regular 8×8 DCT, which is also used in MPEG in a “frame DCT” mode. The 248-DCT is intended to code those DCT blocks with relatively large “intra-frame” motion, that is to say, motion between the two interlaced fields that make up a video frame. This mode still uses an 8×8 DCT block, but each b
Bushmitch Dennis
Lin Wanrong
Le Vu
Matsushita Electric Corporation of America
RatnerPrestia
LandOfFree
Bit number prediction for VLC coded DCT coefficients and its... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Bit number prediction for VLC coded DCT coefficients and its..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Bit number prediction for VLC coded DCT coefficients and its... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3103008