基于第二代小波的图像与视频压缩的研究

【作者】 郭晶磊；

【导师】 韩其睿； 张兴会；

【作者基本信息】 天津工业大学 ， 计算机应用技术， 2008， 硕士

【摘要】 图像和视频是人类可以利用的最主要的信息载体。互联网的发展和多媒体的进步使得多媒体的各种新的应用和服务成为可能,尽管互联网的带宽和设备的存储容量都有所提高,但是数字化了的视频和图像信号的数据量之大是惊人的。数据压缩技术作为解决这一问题的有效途径,越来越受到大家的重视。本文的滤波器采用Daubechies双正交滤波器,在图像压缩中,它获得了比正交小波更好的效果。具体采用的是Daub9/7小波变换。压缩编码采用了分层树集分裂算法（Set Partitioning In Hierarchical Trees SPIHT）以及3D-SPIHT算法。具体描述了嵌入式零树小波EZW（Embedded Zerotree Wavelet）算法与SPIHT算法在压缩中的应用,并且编码实现了SPIHT图像压缩与解压缩算法和3D-SPIHT视频压缩与解压缩算法。将人眼视觉特性引入小波图像编码算法中,并对其进行了改进。本文使用了两种算法来引入人眼视觉特性,其核心是通过两个阈值T₁、T₂将小波图像分成的纹理块、平滑块、边缘块,并且对相应的小波系数赋予不同视觉权值。在第一种算法中,本文提出了新的广义高斯参数估计,引入了误差反向传播（Error BackPropagation BP）神经网络进行参数估计;在第二种算法中,本文将T₁规定为熵值的平均值。而T₂是一个较小的图像块方差。并在三种图像块分界处我们引入了模糊算法。使图像块的划分更加合理。两种算法都取得了较好的实验结果。更多还原

【Abstract】 Image and video are among the most important information carriers for human being. Recently, the expansion of Internet and the development of multimedia technology have made it possible to implement new applications and services of multimedia. Although the Internet bandwidth and storage capability increased dramatically in the past decades, digitalized image and video signals still great challenges to our technology. As a feasible solution, data compression technology has been paid more and more attentions.In this paper, we use Daubechies orthogonal filter to obtain better image compression results than orthogonal wavelet. In detail, we use the Daub 9/7 wavelet converter along with Set Partitioning In Hierarchical Trees （SPIHT） and 3D-SPIHT algorithms. Describe the application of Embedded Zerotree Wavelet （EZW） algorithm and SPIHT algorithm in real world compression tasks. In practice, we implemented the SPIHT image compression algorithm and the 3D-SPIHT video compressing and decompressing algorithm.Introduce improved human visual characteristics into wavelet image coding algorithm. We use two algorithms to introduce human visual characteristics, the basic idea is to separate images into basic detailed or textured regions, smooth regions and strong regions by two threshold values T₁ and T₂ and then give different image parameter to corresponding wavelet coefficient. In the first algorithm, we introduce the BP neural network for the parameter estimation in Generalized Gaussian Distribution; In the second algorithm, we define T₁ the average value of all entropys and T₂ a smaller image deviation. We also use fuzzy algorithm in the dividing position of three different image blocks to make the division of image blocks more reasonable. Both of experiments show very inspiring results.更多还原