【作者】 郝志成；

【导师】 赵继印；

【作者基本信息】 吉林大学 ， 通信与信息系统， 2004， 硕士

【摘要】 1. 引言图像采集和预处理是图像处理领域最重要技术之一。图像采集的目的是把现实世界的模拟图像采到处理通道中并转化成数字图像，这一过程的关键在于如何快速实时地把模拟图像转化成数字图像；而预处理的目的是通过对图像的去噪处理来提高信噪比，突出图像的期望特征。图像去噪方法有时域和频域两类方法，但归根到底都是利用噪声和信号在频域上的分布不同进行去噪的。信号主要分布在低频区域，而噪声主要分布在高频区域，同时图像的细节也分布在高频区域。所以，图像信号去噪中存在一个一直尚未得到很好解决的困难，即在去除图像随机噪声的同时引进图像边缘的模糊，在保留和增强图像边缘的同时又增大了图像的噪声。传统的低通滤波方法将图像的高频部分滤除，虽然能达到降低噪声的效果，但会破坏图像的细节。如何构造一种既能够降低图像噪声，又能够保持图像细节的去噪方法，一直是图像处理领域中理论和实践工作者的研究目标。小波变换这一强有力的信号分析工具的出现，使上述目标的实现成为可能。小波理论是近年来新兴的一门应用理论，并得到了非常迅速的发展，而且由于其具备良好的时频特性，因而实际应用也非常广泛。传统的傅里叶变换有明显的缺陷——无时间局部信息。而实际信号往往是时变信号、非平稳过程，了解它们的局部特性是很重要的。小波本身的特性使它特别适用于分析非平稳信号。小波分析的应用研究是与小波分析的理论研究紧密地结合在一起的。同时，在小波领域又能够划分出很多的分支领域，多进制小波、小波包基、分解尺度、抽取方法等等。可见，在小波不仅功能强大，还有很多亟待解决的问题在等着我们去发现去研究。2. 图像采集系统硬件设计作者首先考察了当前图像采集系统的系统结构，分析了各种系统结构的优缺点，在此基础上作者设计了一种快速、实时的图像采集系统。本系统采用了PHILIPS公司地高速视频解码芯片SAA7111进行A/D转换，以Altera公司的大规模集成FPGA芯片EP1K10TC144作为图像采集控制模块，数据处理芯片<WP=76>选用了TI公司的TMS320C6204。同时，系统中设置了两片大容量的SRAM作为帧存储器，并采用乒乓读写机制以提高存储器的读写速度。3. 去噪算法本文工作的重点集中在图像去噪算法的研究，介绍了当前信号处理领域尤其是图像处理领域去噪算法的发展状况，通过探讨时域和频域的图像去噪方法，进而确定了本文在小波域进行信号处理的基本方针。在众多小波去噪算法中，阈值萎缩法是最常用方法，利用阈值法进行信号去噪处理很大程度上取决于阈值门限和阈值函数的选取。所以又考察了小波去噪算法中对阈值门限和阈值函数的选取理论。阈值法尽管对有些图像的去噪效果好，但从普遍意义上，去噪效果要差一些，在应用上有一定的局限性。因此，开始寻找去噪的新方法，通过对比例萎缩法去噪原理和去噪实例的分析，发现比例法恰好能弥补阈值法的缺点，作者在此基础上提出一种新的小波去噪算法——PCT法，实验证明其适应性和效果上要远远优于阈值法及其改进方法。作者从以上几个不同的角度入手，确立了本文基本的图像去噪研究思想：噪声-小波-阈值法-比例萎缩法-PCT法去噪。针对小波变换的紧支性和去相关性等特性，去噪算法将从噪声干扰中恢复真实的信号。其去噪主要步骤如下：第一步，对图像信号进行三级离散小波变换，得到小波变换系数，其中，k表示分解层数；i，j表示象素点位置。第二步，估计信号小波系数的方差，其计算式为 第三步，利用第二步得到的来计算PCT法中的阈值门限，其计算式为。第四步，运用PCT法的函数式，求出处理后的小波系数，其函数式表示为第五步，对进行离散小波反变换，从而得到去噪后的信号。 <WP=77>4. 结论通过以上几个方面的研究工作，重点提出了小波变换快速算法和PCT去噪算法。在计算机仿真实验中，对Woman图像加噪信号进行处理，并且同常用的硬阈值、软阈值法，硬、软阈值折衷法，导数连续的阈值法及比例萎缩法的结果比较，结果比较，通过对去噪前后的峰值信噪比(PSNR)和均方误差(MSE)的变化情况说明这种去噪算法对于图像噪声的去除是有效的，尤其是在低信噪比时效果更为明显。更多还原

【Abstract】 1. IntroductionThe gathering and the pretreatment of image is one of the most important technology of image process fields. The purpose that the image is gathered is to transforms the analogical image of the real world into a digital image, the key to this course lies in how to transform the analogical image into a digital image fast. And the purpose of the pretreatment is to improve SNR to stress the image characteristic through denoising processing. Image denoising includes space (time) and frequency domain methods. But their essence is to use the differences between noise and signal distributing in frequency domain, the signal is mainly distributing in low frequency part, while noise and image detail mainly in high frequency part. So there always is a difficulty needs to be solved, it’s to eliminate random noise but not causing image edge blurring, to keep and increase image edge but not enlarge noise. The traditional low-pass filter method can restrain the high frequency part and decrease noise but may destroy image detail, and other methods also have other defects. How to construct a denoising algorithm which decreasing noise and keep image detail is always research object of scientists in image processing world. After the appearance of new signal processing tool—wavelet, the object can be implemented easily in reality. Wavelet analysis is a rapid developed new area of nowadays applied mathematics. Wavelet transform(WT) is space(time) and frequency local transform compared to Fourier transform and Windowed Fourier transform(Gabor transform). So it can pick-up local information from signal and can realize multi-analysis to signal by stretching 、 shrinking and shift etc. It solved many problems which can’t be solved by Fourier transform. So we call wavelet transform as “mathematics microscope”. Now wavelet transform is used widely in many areas. Especially in image processing area it shows great advantage and has promising future.2. The Hardware Design of Image Gathering SystemIn this paper, the author studied the structure of the picture gathering system at first, analysed the merit and demerit of different system. On this foundation, the author designed a <WP=79>real-timely picture gathering system. This system adopts the PHILIPS Company’s high-speed video decode chip SAA7111 to perform A/D transition, used the integrated FPGA chip EP1K10TC144 of Altera Company as controlling the picture gathering, and the TMS320C6204 of TI Company was selected for data processing chip. Meanwhile, two SRAM of large capacity were used as the frame memories, and author adopted the ping-pong reading and writing mechanism to improving the speed of reading and writing memories. 3. Denoising AlgorithmThis emphasis of this paper’s job concentrate on the research of image denoising algorithm. Author recommend the state of development of the denoising algorithm at present in signal processing field, especially image processing field. After the research of denoising algorithms in space(time) and frequency domain, the paper proposed a denoising algorithm in wavelet domain. Of all wavelet denoising algorithms, threshold-method is used the most usually. Threshold-method depends on the selection of threshold value and threshold function at great extent. The author studied the theories of selection of threshold value and threshold function in the course. Though threshold method can denoise effectually to some images, from universal significance, it can not do well, and have some limitation on using. So author looked for new denoising method, and found that proportion-shrinking method can remedy the demerit of threshold method through analysis to principle and examples of proportion-shrinking denoising algorithm. On it a new wavelet denoising algorithm--PCT was proposed by author, the experiment conclusion proves its adaptability and denoising effect far superior to threshold method and improved threshold method. The author established the base denoising research thought from different angles discussed above: noise-wavelet更多还原