A Decision Criterion to Relocate Codewords for Adaptive Vector Quantization

H. Takizawa (Japan)


adaptive vector quantization, neural networks, non stationary signal source, equidistortion principle


In many adaptive vector quantization (AVQ) algorithms, codeword relocation is employed for quick adaptation to changes in non-stationary source statistics. However, those algorithms generally determine codeword relocation us ing heuristically predetermined parameters as thresholds; thereby, their performance seriously depends on parame ters. This paper proposes a valid decision criterion whose threshold is automatically adjusted according to optimality of a codebook assessed from the viewpoint of the equidis tortion condition. In this paper, an AVQ algorithm is also presented to verify the criterion validity. Experimental re sults show that the proposed AVQ algorithm can minimize expected distortion and also quickly adapt to changing non stationary signal sources by properly relocating codewords. Accordingly, validity of the proposed criterion is clarified via experimental results.

Important Links:

Go Back