In this report, we introduce a novel feed-forward adaptive quantization scheme called SAPQ (sampleadaptive product quantizer) as a structurally constrained vector quantizer. SAPQ is based on a concept of adaptive quantization to the varying samples of the source and is very different, from traditional adaptation techniques for non-stationary sources. SAPQ quantizes each source sample using a sequence of quantizers. Even when using scalar quantization in SAPQ, we can achieve perforrnance comparable to vector quantization (with the complexity still of the order of scalar quantization). We also show that important lattice based vector quantizers can be constructed using scalar quantization in SAPQ with several examples. We asymptotically analyze SAPQ and propose a simple algorithm to implement it. We numerically study SAPQ for independent, and identically distributed Gaussian and I;aplacian sources. Through our numerical study, we find that SAPQ using scalar quantizers achieves typical gains of 1 3 dB in distortions over the Lloyd-Max quantizer. By employing SAPQ, we have extensively conducted image compressions. We considered a uniform quantizer for the current H.263 standard and a nonuniform quantizer for the differential pulse code modulation for images. We also show that a generalized SAPQ can be used in conjunction with vector quantizers to further improve the gains, especially for high correlated image signals at large vector dimensions. Finally, the error-resilient aspect of SAPQ is also investigated. We find that SAPQ is robust to the channel noise, while achieving gains.
Date of this Version