图像小波域视觉重要性顺序模型比特零树预测熵编码

Image Wavelet Transform Entropy Coding Using the Visual Importance Order Model and Bit Zerotrees Predicting

本文研究一种静态图像小波（Ｗａｖｅｌｅｔ）域系数压缩方法，可以在任意给定比特数时获得一定视觉意义下的最佳图像质量。该方法根据图像小波分解和人类视觉（ＨＶＳ）的特点及其关系，对系数进行不同间隔的量化，并规定系数的视觉重要性顺序，结合零树数据结构，对系数进行该顺序的比特层零树扫描和预测，输出符号数据流，最后用自适应算术熵编码实现高效率编码。该方法易于用ＶＬＳＩ实现，输出比特率任意可调。计算机模拟结果显示，对于图像（Ｌｅｎａ２５６）在０．２～０．

In this paper, we presend a novel still image coding based on wavelet transform. The 2D wavelet transform decomposes an image into both spatial and spectral sub images coefficients. The decomposition closely mimic the human visual system(HVS) and there is very strong correlativity of coefficients among those sub images. Thus, in this method linear quantizer with step size derived from a HVS model is first applized to each coefficient and then the coefficient's visual importance order to reconstruct image is set up. By bit zerotree structure, the coefficients are scaned in that order and encoded by an adaptive arithmetic coder. The algorithm can be mapped easily onto VSLI. For Lena256 standard monochrome image, it gives high acceptable quality reconstructed images at 0.2￣0.3bit/pixel.