QQ客服 官方微博 反馈留言

题名	基于小波分析的静止及运动图像压缩研究
作者	向阳朝
学位类别	工学博士
答辩日期	1996-06-01
授予单位	中国科学院自动化研究所
授予地点	中国科学院自动化研究所
导师	戴汝为
关键词	图像压缩 小波分析 矢量量化 时变小波 运动补偿 网格匹配 图像分割 image coding wavelet analysis vector quantization time varying wavelet motion compensation mesh matching image segmentation
学位专业	模式识别与智能系统
中文摘要	图像通信是多媒体通信中的一个重要组成部分，目前的图像压缩方法却仍不能很好 地满足正日益拥挤的网络带宽的要求。而小波分析为进一步提高压缩比提供了新的数学工 具。作为中科院自动化所人工智能实验室的一项863课题一“智能多媒体通信系统”中的 一部分，作者在静止以及运动图像的小波压缩方面做了大量工作，主要研究内容包括： 1． 提出了基于网格自适应特征跟踪的小波压缩方法。用可分离二维小波对图像进行小波 变换时，高通滤波器具有水平、竖直和对角三个方向的选择性，因而对其它方向的高 频信息(即边缘)的压缩效果不好，在高压缩比的情况下，产生明显的边缘Gibbs现 象(即振铃现象)。与一般小波变换压缩方法相比，本文所提出的压缩方法增加了两个预 处理过程，即：三角形网格特征跟踪和空间坐标变换。首先利用三角形网格对图像进 行边缘特征跟踪，使网格点跟踪最近的边缘特征点，然后根据网格点的位置使图像产生 形变f即对之进行坐标变换)，使原图像中的边缘方向分别大致变成水平、竖直或对角方 向，以适应只具有这三个方向高通的滤波器。这一方法有效地减轻了Gibbs现象，在 保持图像客观压缩质量(PSNR)的同时，提高了主观压缩质量；并通过确定网格点的特 征点匹配邻域以及引入弹性网的概念，使得网格特征跟踪算法的计算量非常小。 2． 由小波理论可知，具有较长支撑集的小波能有效地集中图像的能量，而具有较短支撑 集的小波的空间分辨率高。这意味着，具有较长支撑集的小波适合对灰度变化较平坦 的图像变换编码，具有较短支撑集的小波适合对细节较多的图像变换编码。如果在一 幅图像中，我们能根据边缘情况自适应地改变小波支撑集的长度，解压缩图像中边缘 附近的Gibbs现象将有可能大量减少并同时提高编码效率。滤波器自适应变化所带来的 主要问题是在滤波器变化的过渡期间如何准确重构原信号。本文刚两个双线性正交小 波滤波器组构造时变小波变换，给出了两个不同支撑集的小波滤波器在自适应切换时， 过渡期间象素值精确重构的公式。同时，通过在小波分解过程中引入小波系数的反馈， 实现了反馈时变小波变换，使小波滤波器组白适应切换的位置信息被隐含在小波系数 编码中，而不必再作为附加信息存储。 3． 在图像的灰度平坦区，图像小波系数的绝对值较小，而在图像边缘附近，小波系数的 绝对值很大。针对这一特点，本文提山了一
英文摘要	Image communication is a key part of multimedia communication, but the present image compression method is still not good enough to fit for the network bandwidth which is becoming narrower. While new mathematical tool- wavelet analysis- makes it possible to improve compression ratio highly. As one part of "863" project --"Intelligent Multimedia Communication System", a lot of work has been done in the field of still and sequence image compression. The main research contains: 1. An adaptive mesh feature tracking method was proposed to alleviate the ring effect in the margin of edge(i.e. Gibbs phenomenon). When coding with separate wavelet, three orientation (horizontal, vertical and diagonal) are preferred in high pass filter, while other orientations are not coded effectively, this will lead to serious Gibbs phenomenon especially in high compression ratio. This paper utilizes feature tracking mesh to solve such problem. First a triangle mesh is used to track the image edge orientation in every local area, all mesh grids is set as close as possible to edge, then image is warped according to the position of these mesh grids. After warping, all edge orientations are almost near to one of the three preferred orientations. The subjective quality of reconstructed image is improved while objective quality is almost the same. 2. A class of time varying wavelet transform was proposed. Wavelet with longer support length is used in the smooth area of image to integrate energy effectively, while wavelet with shorter length is used in the vicinity of edges. These two sets of wavelet transform switch automatically according to the pixel values of image. The main problem for the design of time-varying filter is how to reconstruct exactly the signal during the transition period. An exact reconstruction method between any two sets of the widely used biorthogonal wavelet bases was proposed in this paper. And even more, by feedback of quantized wavelet coefficients, side information is embedded in the coding bit stream. 3. In the smooth area of image, the wavelet coefficients are very small, while in the vicinity of image edge, they are relatively very large. According to this property, a new quantization algorithm combining scalar and vector quantization method was presented in this dissertation. Experiments showed that not only compression ration is improved against scalar quantization, but also distinct image edge is preserved. 4. With discrete wavelet transformation(DWT) and mesh matching(MM), a new multi-resolution video coding scheme is presented in this dissertation. DWT is performed both in intraframe and displace frame difference(DFD) instead of DCT in MPEG standard to remove the signal correlation. Mesh matching is used to obtain high quality of motion compensation. Si
语种	中文
其他标识符	369
内容类型	学位论文
源URL	[http://ir.ia.ac.cn/handle/173211/5663]
专题	毕业生_博士学位论文
推荐引用方式 GB/T 7714	向阳朝. 基于小波分析的静止及运动图像压缩研究[D]. 中国科学院自动化研究所. 中国科学院自动化研究所. 1996.

个性服务

查看访问统计

相关权益政策

暂无数据

收藏/分享

除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。