Predictive vector quantization based upon first-order Markov model for image transmission

Abstract

One major problem with nonadaptive image vector quantization is that the edge cannot be reconstructed efficiently. A predictive vector quantization scheme based on a first-order Markov model is presented to address this problem. At the heart of this design is the decomposition of the original image source in the spatial domain into two, three, or four interleaved subimages. The subimages can be coded separately using vector quantization. Since the first-order pixel-to-pixel correlation is usually very high in ordinary images, the subimages are very strongly correlated. Only one or two subimages can be directly coded by vector quantization. The remaining subimages are calculated by linear prediction, and an activity index is applied to the predictive errors. At low activity or shade areas, the error is lower than a preset threshold value and is assumed to be zero. At high activity or edge areas, the error is greater than the preset threshold value and is coded by vector quantization. Therefore, the bits saved at low-activity areas are allocated to the high-activity areas. This results in good edge reproduction and efficient prediction. >