Image Compression Using Vector Quantization with Variable Block Size Division

抄録

In this paper, we propose a method for compressing a still image using vector quantization (VQ). Local fractal dimension (LFD) is computed to divided an image into variable block size. The LFD shows the complexity of local regions of an image, so that a region of an image that shows higher LFD values than those of other regions is partitioned into small blocks of pixels, while a region of an image that shows lower LFD values than those of other regions is partitioned into large blocks. Furthermore, we developed a division and merging algorithm to decrease the number of blocks to encode. This results in improvement of compression rate. We construct code books for respective blocks sizes. To encode an image, a block of pixels is transformed by discrete cosine transform (DCT) and the closest vector is chosen from the code book (CB). In decoding, the code vector corresponding to the index is selected from the CB and then the code vector is transformed by inverse DCT to reconstruct a block of pixels. Computational experiments were carried out to show the effectiveness of the proposed method. Performance of the proposed method is slightly better than that of JPEG. In the case of learning images to construct a CB being different from test images, the compression rate is comparable to compression rates of methods proposed so far, while image quality evaluated by NPIQM (normalized perceptual image quality measure) is almost the highest step. The results show that the proposed method is effective for still image compression.