Color Image Compression with Vector Quantization

We propose vector quantization (VQ) with variable block size for color image compression. Variable block size for VQ has so far been implemented using quad-tree decomposition. This method focus on homogeneity of local regions of an image. We consider the complex regions of an image to be more essential than homogeneous regions because complex regions may have a wealth of information. We propose variable block size for VQ based on local fractal dimensions (LFDs) and compressed color image. Encoded images obtained by using the proposed method were evaluated by both compression rate and image quality. Results of computational experiments show that the proposed method is superior to conventional VQ with fixed block size.