Object-based Still Image Compression Study

The research and application results show that image compression has been one of the most important technology for network transmission in last decades. Moreover, object-based coding is a kind of up-and-coming technology of image compression. At present, JPEG2000 standard and MPEG-4 standard adopt the technology. This paper pays attention to still image compression with an eye to object-based coding technology.JPEG2000 supports two ROI coding methods: the Maxshift method and the general scaling based method. The Maxshift method has the disadvantage of having to code all ROI regions prior to coding any part of the non-ROI region. On the other hand, the general scaling based method of JPEG2000 only supports rectangle and ellipse ROIs.This paper designs an effective project of object-based still image encoder and decoder according to the idea of the ROI coding of JPEG2000. The project includes three parts:(1) The project segments and extracts the region of interest with image segmentation technology and presents an effective method of contour extraction with human being as region of interest. The method confirms human being location in the image with face detection method and accurately segments the image with edge detection and region growing method. Lastly, it extracts the image region of interest with human being model.(2) The project codes the human being contour with the contour coding technology of arbitrary shape object. This paper presents a simple and effective shape-coding algorithm of arbitrary shape ROI -VEDCC. The proposed algorithm supports multiple arbitrary shape ROIs coding, and can control the relative importance between ROIs and BG. The VEDCC is an efficient and less complex shape-coding algorithm.(3) The project compresses the segmented human being and background in term of different compression rate with texture coding technology of arbitrary shape object. The paper adopts region of interest coding method of JPEG2000. But the pixel values of ROI boundary of decoding image distort very large because of the shifting of background wavelet coefficients. For this problem, this paper takes the place of JPEG2000 wavelet transforms with Shape-Adaptive Discrete Wavelet Transforms. The method largely improves the image quality of ROI boundary.Experiment results illustrate the performance of the proposed project.