Power efficient vector quantization design using pixel truncation

Abstract

Vector quantization image encoding requires a huge amount of computation and thus of power consumption. In this paper a novel method is proposed for the reduction of the power consumption of vector quantization image processing by truncating the least significant bits of the image pixels and the codewords elements during the nearest neighbor computation. Experimental results prove that at least 3 pixels/elements bits can be truncated without affecting the picture quality. This results in an average 65% reduction of bus power consumption and in an average 62% reduction of the power consumed in major data path blocks.