Adaptive resolution vector quantization technique and basic codebook design method for compound image compression

In order to increase the performance of image compression by vector quantization (VQ), we propose a systematic codebook design method without using learning sequences for 4/spl times/4 and 2/spl times/2 pixel blocks. According to the method, the codebook can be applied to all kinds of images and exhibits equivalent compression performance to the specific codebooks created individually by conventional learning method using corresponding images. Furthermore, we have developed a novel VQ-based image-coding algorithm suitable for compound images. Adaptive resolution VQ (AR-VQ) method, which is composed of three key techniques, i.e., the edge detection, the resolution conversion, and the entropy coding, can realize much superior compression performance than the JPEG and the JPEG-2000. On the compression of the XGA (1024/spl times/768 pixels) images including text, for instance, there exist an overwhelming performance difference of 5 to 40 dB in compressed image quality.