Date post: | 21-Dec-2015 |
Category: |
Documents |
View: | 219 times |
Download: | 0 times |
A New Dynamic Finite-State Vector Quantization
Algorithm for Image Compression
Jyi-Chang Tsai, Chaur-Heh Hsieh, and Te-Cheng Hsu
IEEE TRANSACTIONS ON IMAGE PROCESSINIG , NOVEMBER 2000
VQ for image coding
• VQ which exploits the correlation among neighboring blocks– Predictive VQ– Finite-state VQ (FSVQ)– Dynamic FSVQ– Address VQ– Index search VQ
Vector Quantization (VQ)
X1
X2
DFSVQ
Proposed DFSVQ
• Search the best block in predefined search area which contains previously encoded data.
• The current input block can be represented by the best block, dynamic codebook or super-codebook.
• The search for the the best block from the the search area is equivalent to expanding the code-vector space. Thus the picture is superior to the basic VQ with full search method.
Proposed DFSVQ (cont.)
Simulation Results
VQ
0.563 bpp,
31.10 dB
DFSVQ-N
(0.430 bpp. 31.06 dB),
Original
SMVQ
(0.412 bpp, 31.10 dB),
PDFSVQ
0.246 bpp, 31.07 dB
Conclusions
• For each input block, the PDFSVQ first searches the best block. Then, the current block is encoded by the best block, dynamic codebook or super-codebook, depending on the coding distortion.
• The PDFSVQ exploits the global correlation of image blocks rather than local correlation in conventional memory VQs.
Conclusions (cont.)
• The PDFSVQ expands the codebook space without extra overhead information bits; thus, it achieves better rate-dis-tortion performance and visual quality than conventional DFSVQs.