1996 年 33 巻 1 号 p. 46-56
This paper deals with the problem of compression of high-resolution images using the discrete cosine transform (DCT). In part 1 of this paper, we have proposed a new method for compressing high-resolution images using DCT with variable block sizes. Comparing the block sizes of 4×4, 8×8 and 16×16, it is shown that the largest block size is not always the best. In particular, 4×4 will often be more useful than others for obtaining high-quality reconstructed images. The value of adjacent difference is shown to be very effective for the determination of block size. Here, we propose a method for the compression with variable block sizes based on the visual characteristics. A visual sensitivity function is obtained in the spatial frequency domain. The concept of weighted signal-to-noise ratio (WSNR) is introduced using the visual sensitivity function. The WSNR is used instead of SNR for evaluating the reconstructed images. As high-resolution images, 6 pieces of SCID (Standard Colour Image Data) images are used for simulation. Various kinds of simulation results using the WSNR are presented, and they are compared with those using the conventional SNR. These results show that the WSNR is more useful for the compression of high-resolution images. Finally, a new quantization table using the visual characteristics is presented, and the proposed method is extended to high-resolution colour images.