Improved Distortion Measures for Image Vector Quantization
Anamitra Makur · 2005
The role played by the distortion measure in a vector quantization image encoder is very important. In the following paper we suggest a general class of distortion function, the input-dependent weighted square error distortion, which is computationally simple and can be used to incorporate some psychovisual characteristics. Two techniques of using this distortion function have been discussed. Incorporating various classes depending on image activity in the distortion measure, better subjective quality can be achieved without added complexity, as our simulation studies have shown. This strategy of classified codebook is optimal and more general than the conventional subcodebook method. We also show that incorporating dimensional emphasis, blocklness can of marginally more computation.