【作者】 侯颖；

【导师】 李国民；

【作者基本信息】 西安科技学院 ， 通信与信息系统， 2002， 硕士

【摘要】 图像是多媒体应用中的重要媒体，而传输频带和存储空间的限制，使得图像压缩技术成为当前信息传输系统中最关键的技术之一。小波变换技术以其良好的空间-频率局部特性和与人眼视觉特性相符的变换机制，在图像编码领域获得了广泛的应用和研究。 针对静止图像的压缩编码，在嵌入式小波零树编码方法的基础上，本文提出了一种具有分块预处理的改进新算法——嵌入式小波零树分块图像编码算法。该方法首先将图像经过小波变换后产生不同分辨率下的各子带图像；然后在阈值T的标准下，对图像水平、垂直和对角线方向的高频子带系数进行四叉树图像分割预处理，从而产生分块类型编码；最后再对均匀图像块进行嵌入式小波零树编码，产生系数类型编码和幅值编码。这种改进的编码算法使得大量非重要系数集中成图像块表示，更好地利用小波变换后的系数，有效地降低码率，同时增强图像压缩后恢复图像的视觉效果。新算法具有与嵌入式小波零树编码相同特点，即可以产生嵌入式码流、不需要训练码本、且在所要求的精度下可以随时结束编码。实验及仿真结果表明，这种改进算法得到的重构图像其主观视觉效果良好，与嵌入式小波零树算法相比，峰值信噪比(PSNR)在相同码率的情况下有较大的提高(平均PSNR值约提高0.37—0.46dB)。更多还原

【Abstract】 Image is very important medium at application of multimedia. Because of limitation of transmission bandwidth and storage capacity, so image compression became one of critical techniques at information transmission system. In recent decade, the technique of wavelet transform has been applied and studied extensively because of its good features of time-frequency and human visual system.In this paper, a improved low bit rate image compression algorithm, namely embedded wavelet zero-tree segmntation-based image coding, is put forward about still images based on embedded wavelet zero-tree image coding scheme. First, image is divided subband images of various scales by wavelet transforming. Then at this algorithm the coefficient in horizontal vertical and diagonal orientation sub-images of all scales are quadtree segmented at threshold T criterion. At last these uniform blocks are encoded based on EZW. This improved coding algorithm not only is with the property of EZW but also let unimportant coefficient becoming image blocks to represent so that we can effectively apply the wavelet coefficients of image, low bit-ratio and intensify the visual effective of reconstruction images at the same time. To the embedded zerotree wavelet algorithm(EZW) this paper’s coding algorithm is similar the properties which are yielding a fully embedded code, terminating the encoding and decoding at any point respectively under requirement of rate-constrained or distortion-constrained, absolutely no training and no pre-stored codebooks. Through the experimental and simulated results, we can demonstrate that the reconstruction images are good, and the peak signal-to-noise ratio (PSNR) of the reconstructed image is improved (mean improved PSNR is about 0.37-0.46dB ) compared to embedded zero-tree wavelet encoding (EZW) algorithm at the same bit-ratio.更多还原