Efficient vector quantization for color image encoding

The practical application of adaptive vector quantization to the encoding of a database of large color images is demonstrated. Methods for search space reduction, improved codebook initialization, empty cell treatment, and color vector coding are presented, with particular emphasis on encode time reduction. These variations to the basic vector quantization algorithm are compared in terms of computational complexity, execution time, and image quality. Experimental results are presented for a typical 256 × 256 pixel color image at a representative data rate of 2.0 bits/pixel, demonstrating over two orders of magnitude improvement in computation time.