A conditional entropy-coded multi-stage vector quantizer for image coding

Abstract

Summary form only given. A vector quantization (VQ) algorithm that utilizes this interblock correlation is proposed. Despite its usefulness for low bit-rate applications, the standard, table-lookup VQ has the drawback of exponential growth of both storage and computation requirements with vector dimension. In practice, the vector size is typically from 3*3 to 5*5. With this size, it has been found that there is still strong correlation between neighboring blocks. Experiments have shown that the approach generally gives better coding performance than previously designed VQ codes. So far, experimental results for coding various digital images are in the range of 30 dB to 36 dB with about 0.5 b.p.p.<>