Distortion/decoding time tradeoffs in software DCT-based image coding

We present a general framework for variable complexity algorithms (VCA) and study the related issue of defining a minimum average complexity implementation. As an example we consider implementations of the inverse DCT (IDCT) which minimize the average computation time by taking advantage of the sparseness of the quantized input data. Since the decoding speed depends on the number of zeros in the input we then present a formulation that enables the encoder to optimize its quantizer selection so as to meet a prescribed "decoding time budget". This leads to a complexity-distortion optimization technique which is analogous to well known techniques for rate-distortion optimization. In our experiments we demonstrate significant reductions in decoding time.