Generalized product codes: a framework for the design of structured vector quantizers

Coders for audio and visual sources can be viewed as structured vector quantizers. The authors propose a generalized product code (GPC) model for a class of structured source coders. The model serves as a basis for exploring design approaches and algorithms applicable to the class at large. These demonstrate the generality of the GPC model through examples and show how encoding and storage complexity can be independently traded off with distortion performance. An approach for joint codebook optimization is discussed which is a performance enhancement step incurring no run-time cost. By viewing the entire coder as a product code or composition of product codes, the complexity of individual coder components is incorporated directly into the exploration of the overall system performance-complexity tradeoff