Pulse or digital communications – Bandwidth reduction or expansion – Television or motion video signal
Patent
1995-01-17
1997-11-25
Chin, Stephen
Pulse or digital communications
Bandwidth reduction or expansion
Television or motion video signal
348403, 348417, 348420, 348422, 382253, H04B 166
Patent
active
056920121
DESCRIPTION:
BRIEF SUMMARY
FIELD OF THE INVENTION
The invention relates to the method defined in the introductory section of patent claim 1 for image compression coding in an image transmission system, particularly for realizing compression coding in a digital image transmission system.
BACKGROUND OF THE INVENTION
In the prior art there is known a vector quantization (VQ) method for image compression coding. The VQ method is discussed extensively in the reference publication /1/. In the VQ process, the image to be transmitted is divided into blocks, and for each block, there is searched in the codebook a codevector that best corresponds to the particular block. The vector index of this codevector, i.e. its location in the codebook, is transmitted. At the receiving end, the said vector is looked up in a corresponding codebook on the basis of the vector index of the codevector. The transmitted image is reconstructed, block by block, by means of the said codevectors. The codebook is designed beforehand by using training images and a suitable codebook design algorithm. In a corresponding fashion, the training images are divided down to the used block size while creating the codebook.
A drawback of the said VQ method is its limited block size. The efficiency of the process stems from the fact that all statistical dependencies within the block are taken into account. The correlations over the block boundaries remain unexploited. In standard vector quantization, large scale correlations can be utilized to achieve a higher compression only by using a larger block size. This unfortunately leads to an exponential growth of the codebook size for a fixed picture quality. The number of comparisons to find the best codevector in the codebook also grows respectively. Therefore VQ becomes computationally unpractical with large block sizes. Usually the block size is limited to the order of 4.times.4 pixels, which leaves a lot of statistical dependencies beyond the reach of the method.
The problem with large block size in vector quantization is twofold. On the one hand, the sheer size of the codebook becomes a memory storage problem. On the other hand, the computing time to find the best codevector may become excessive. Several applications of the VQ method have been developed to overcome these problems.
In a modified vector quantization process, the tree search VQ (TSVQ), the computing time problem is addressed by using a binary search procedure to find the best codevector. In image quality terms, TSVQ works slightly worse than the standard VQ.
Another modified vector quantization process, the cascaded VQ, represents a direct attack on the main problem of standard VQ, i.e. the problem with block size. In this method, a relatively large block size is chosen in order to capture the large scale correlations. However, the employed codebooks are reasonable in size and thus easily manageable. To compensate for the unavoidable lack of accuracy, the method is applied in successive coding stages. At each of the later stages, the coding error of the previous stage is quantized. A drawback is that any correlations that may exist between successive codevectors are lost.
In a third modified vector quantization process, which is a refinement of the cascaded VQ, the block size is changed when proceeding from the first coding stage to the next. The coding error, which contains the fine details of the image, is coded using sub-blocks. By using this method, the number of such codevectors in the codebook that are needed to represent the small scale details can be drastically brought down with a small block size.
In a fourth modified vector quantization process, the classified VQ, the blocks are classified prior to coding. The classification is based on local image features. This improves especially the representation of edges, which is found to be rather poor in standard VQ. For each block class there is a separate codebook. A reasonably small codebook will suffice since the image blocks in the same class have more or less similar character. Inter block correlatio
REFERENCES:
patent: 5468069 (1995-11-01), Prasanna et al.
patent: 5521988 (1996-05-01), Li et al.
Valli Seppo
Virtamo Jorma
Chin Stephen
Valtion teknillinen tutkimuskeskus
Vo Don
LandOfFree
Method for image compression coding in an image transmission sys 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 for image compression coding in an image transmission sys, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method for image compression coding in an image transmission sys will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2113547