STEREOSCOPIC COLOR IMAGE CODING BASED ON PERCEPTUALLY OPTIMIZEDJPEG2000 CODER

• Main Authors: Tsung-Hao Li, 李宗澔
• Other Authors: Chun-Hsien Chou
• Format: Others
• Language: en_US
• Published: 2005
• Online Access: http://ndltd.ncl.edu.tw/handle/37585987093534922127

碩士 === 大同大學 === 通訊工程研究所 === 93 === Through the development of the display technology in these years, the 3-D image system has become more and more popular. The most obvious difference between 3-D and 2-D image is the depth of perception. One of the easiest ways to retrieve a vision with depth is to let the viewer see the left-view image in the image pair with the left eye, and vice versa. Because a stereoscopic image is generated with slight differences in viewing the same scene from different perspectives, lots of stereoscopic visual redundancies actually exist. Therefore, a good stereoscopic algorithm should be able to remove various kinds of redundancies. In this thesis, we present a stereoscopic color image coder based on perceptually optimized JPEG2000. This algorithm can not only remove the statistical and visual redundancies in single image, but also the stereoscopic visual redundancies that exist inherently in stereo pairs. According to the characteristic of human visual perception of stereo images, we can perceive an acceptable visual quality of the stereo pair as the visual quality of one image in the pair is assured. Therefore, the perceptually optimized JPEG2000 coder is exploited in the left-view image in the stereoscopic pair during researches. Because JPEG2000 is the new ISO/ITU standard for still image compression. The multi-resolution wavelet decomposition and the two-tier coding structure of JPEG2000 fit in incorporating the human visual model into the coding algorithm. The right-view image of the stereo pair is predicted by disparity estimation. With the right-view image and the predicted image, we can generate a residual. Because we only need a decent visual quality, the residual will only encode the vital information which will affect visual quality. Simulation results show that the acceptable visual quality can be obtained at an average bit rate of 0.21 bpp with PSNR and PSPNR equal to 33 dB and 38 dB respectively. The result shows that our proposed algorithm not only has a good visual quality but also improved the coding efficiency.