Adaptive quantization without side information

We propose to extend some of the ideas of adaptive lossless compression to design an adaptive quantization algorithm. Noting that the performance of an arithmetic coder is as good as its estimation of the statistics of the input, we split our quantizer into two building blocks: model estimation and quantizer design. The main idea is that, as long as the model estimation manages to track down the changes in source statistics, a standard quantizer design technique which assumes that the source follows the estimated model can be used. As an example of this type of design we study an adaptive scalar quantization scheme where we impose the restriction that no side information can be sent, i.e. encoder and decoder must perform their adaptation based on the quantized information.<
>