太原理工大学图像处理课程设计报告.doc

图像处理课程设计报告设计题目： 图象恢复研究 专业班级： 学生姓名： 学 号： 摘要卫星遥感图像是卫星遥感器在轨运行的过程中对地面拍摄所成的图像，获取图像需经过大气、遥感器成像、电子信号传输等多个环节，而每个环节均有可能对图像质量造成影响，使图像退化，因此需要对退化的遥感图像进行恢复处理 目前，国内外已有很多关于图像恢复的算法，然而现有的算法大多是基于点扩展函数PSF或退化函数的知识进行的，有的恢复算法还需知道传感器平台参数、成像环境参数等而一般情况下，这些条件是较难得到全部满足的，并且精确模拟的代价也很大在这些情况下，用现有的算法进行恢复往往存在着一定的困难因此，在已知条件较少，只给出退化图像本身信息的情况下，如何有效地恢复图像是本文研究的目的和重点本文是根据MTF的频域下降特性进行图像恢复的 本文的主要工作如下 1、首先，对图像进行预处理根据遥感图像中条带噪声和孤立噪声不同的特点，分别采用了基于傅立叶频谱分析的频域滤波算法和空域局部自适应中值滤波算法进行去噪这两个去噪算法在一定程度上解决了去噪的同时会模糊图像的问题，能够在有效去噪的同时比较好地保持图像非噪声信息不受损失 2、接着，基于调制传递函数MTF进行图像恢复MTF表征了光学系统成像性能的好坏，它是空间频率的函数，随着空间频率的升高而逐渐下降本文正是利用了MTF的这一频域下降特性对图像进行恢复该算法主要分为如下步骤先计算图像的MTF，然后通过MTF的频域下降曲线建立频域退化模型，进而在频域对图像进行恢复，最后逆变换到空域得到恢复后的图像实验表明，本文的算法能够取得一定的恢复效果 3、另外，本文还根据MTF和点扩展函数PSF的转换关系，通过MTF建立PSF模型，并将所建的PSF模型应用到基于PSF的经典图像恢复算法中同时对不同的PSF模型进行了比较实验，以得出合适的PSF实验结果表明，将本文建立的PSF应用到基于PSF的经典恢复算法中，能够在一定程度上恢复图像 本文不仅对遥感图像进行了研究实验，也对其它类型图像进行了同样的实验通过实验发现，本文算法对遥感图像和非遥感图像都能够取得一定的恢复效果但本文算法仍有很大的研究空间，有待进一步的研究AbstractSatellite remote sensing satellite sensor image is in orbit on the ground during the shooting into the image, the image needs to get through the atmosphere, imaging sensors, electronic signal transmission and other links, each link are likely to cause the image quality the impact of the image degradation, so the need to restore degraded image processing of remote sensing is currently at home and abroad has been a lot about the image restoration algorithm, however, existing algorithms are mostly based on point spread function PSF or degradation function of knowledge, there The recovery algorithm needs to know the parameters of the sensor platform, imaging, environmental parameters, etc. In general, these conditions are more difficult to be fully met, and the price is also very accurate simulation in these cases, using existing algorithms tend to recover there are some difficulties, therefore, less in known conditions, degradation of the image itself is given only the information, how to effectively restore the image and purpose of this study is the focus of this paper is based on MTF in the frequency domain characteristics of image recovery down as a major work of this paper, first, the image pre-processing of remote sensing images based on noise and isolate the noise with different characteristics, were used based on the Fourier spectrum analysis of the spatial frequency domain filtering algorithm and the local adaptive median filter algorithm denoising two denoising a certain extent, solve the denoising problem will also blur the image, denoising can be effective while better information to maintain the image noise without loss of non-2, and then, based on the modulation transfer function MTF MTF for image restoration system characterized by optical imaging performance is good or bad, it is a function of spatial frequency, spatial frequency increases with the gradual decline in this article is the use of the MTF of the decreased frequency domain characteristics of the image to restore the algorithm is mainly divided into the following steps to calculate the image of the MTF, then down through the frequency domain of MTF curve degradation model to establish the frequency domain, and thus in the frequency domain image restoration, and finally the inverse transform to obtain the restored image airspace experiments show this article algorithm is able to achieve some recovery effect 3, in addition, this article according to MTF and point spread function PSF of the conversion relationships, through the establishment of PSF model MTF and the PSF model to build the PSF based on the classic image restoration algorithm while various The PSF model experiments were compared to arrive at the appropriate experimental results show that the PSF will be applied to the paper, the PSF PSF-based recovery algorithm in the classic, to a certain extent, this is not only to restore the image of remote sensing images of the research experiments, also other types of images the same experiment was found by the algorithm of remote sensing images and non-remote sensing images are able to achieve a certain effect, but the recovery algorithm is still much room for research, pending further study目 录图像处理课程设计报告1一、设计目的、任务与要求61、设计目的62、设计任务63、设计要求6二、总体方案设计71、图像退化模型72、退化函数估计9三、各功能模块的主要实现程序10四、测试和调试12五、结论与心得14六、参考文献14一、设计目的、任务与要求1、 设计目的1） 熟悉和掌握用MATLAB程序设计方法；2） 熟悉和掌握MATLAB图像处理工具箱；3） 熟悉图像运动失真和复原的基本原理；4） 学会运用MATLAB工具箱对图像进行处理和分析。2、 设计任务1) 自选黑白图像，并参考P.96获得失真图像；2) 对失真图像进行FFT，并从频谱上研究如何获得失真参数。3) 用获得的参数对失真图像加以恢复(参考P.99)。3、 设计要求利用图像恢复的基本原理在MATLAB环境下编写程序，对静态图像进行图像失真和恢复，并观察分析。自选黑白图像，通过图像运动失真函数使图像失真。并对失真图像进行傅立叶变换，观察变换后的频谱图，考虑是否可以从中获得失真参数值。用获得的失真参数值对失真图像进行图像复原，消除运动模糊。最好能通过程序自动从频谱中获取失真参数，在MATLAB中进行调试和运行，并得出结果。二、 总体方案设计1、 图像退化模型1） 退化模型图像复原的关键在于建立图像退化模型。这个退化模型应该能够反映图像退化的原因。由于造成图像退化的因素很多，而且比较复杂，不便于逐个分析和建立模型；因此，通常将退化原因作为线性系统退化的一个因素来对待，从而建立系统退化模型来近似描述图像函数的退化。如图1所示，这是一种简单的通用图像退化模型，它将图像的退化过程模型化为一个退化系统（或退化算子）H。有图1可见，一幅纯净的图像f（x，y）由于通过一个系统H以及引进外来加性噪声n（x，y）而退化为一幅图像g（x，y）。图1 图像退化模型图1的输入和输出具有如下关系：g(x,y)=Hf(x,y)+n(x,y) 如果暂不考虑加性噪声n(x,y)的影响，即令n(x,y)=0,则有g(x,y)=Hf(x,y)下面考察退化系统H的性质。设为常数，则H有以下性质。（1）齐次性Hkf(x,y)=kHf(x,y)=kg(x,y)即系统对常数与任意图像乘积的响应等于常数与该图像的响应的乘积。（2）叠加性即系统对两幅图像之和的响应等于它对两个输入图像的响应之和。（3）线性同时具有齐次性与叠加性的系统就称为线性系统。线性系统有不满足齐次性或叠加性的系统就是非线性系统。显然，线性系统的齐次性与叠加性，为求解多个激励情况下的响应带来很大方便。（4）位置（空间）不变性Hf(x-a,y-b)=g(x-a,y-b)其中a和b分别是空间位置的位移量。这说明图像中任一点通过系统时的响应只取决于该点的输入值，而与该点的位置无关。在图像复原处理中，尽管非线性和空间变化的系统模型更具有普遍性和准确性；但是，它却给处理工作带来了巨大的困难，它常常没有解答或者很难用计算机来处理。因此，在图像复原处理中，往往用线性和空间不变性的系统模型加以近似。这种近似的优点是可以直接利用线性系统中的许多理论与方法来解决图像复原问题。所以图像复原处理，特别是数字图像复原处理主要采用线性的、空间不变的复原技术。2、 退化函数估计图像复原的主要目的是在给定退化图像g（x，y）以及退化函数H和噪声的某种了解或假设时，估计出原始图像f（x，y）。我们必须在进行图像复原前对退化函数进行估计。估计退化函数的方法一般有三种：1） 图像观察估计法。2） 试验估计法。3） 模型估计法。在课程设计中我们用模型估计法实现：匀速直线运动造成的模糊就可以运用数学推导出其退化函数。假设对平面匀速运动的物体采集一幅图像f（x，y），并设和分别是景物在x和y方向的运动分量，T是采集时间长度，忽略其他因素，实际采集到的由于运动造成的模糊图像g（x，y）为：其傅立叶变换为 改变积分顺序，有再利用傅里叶变换的位移性，有 令则可写成我们熟悉的形式：如果给定运动量和，退化传递函数可直接得到。 假设当前图像只在x方向做匀速直线运动，即 由上式可见，当t=T时，f(x,y)在水平方向的移动距离为a。得 上式表明，当n为整数时，H在u=n/a处为零。若允许y分量也变化，且按运动，则退化传递函数成为：三、各功能模块的主要实现程序C=imread(cameraman.tif); %读入原图像subplot(1,2,1);imshow(C); %显示原图像LEN=35;THETA=50;PSF=fspecial(motion,LEN,THETA); %使原图像做长度变换为35，角度变换为50的运动MF=imfilter(C,PSF,circular,conv); %卷积，使图像失真subplot(1,2,2),imshow(MF);imwrite(MF,cameraman-MF.tif); %保存失真图像I=imread(cameraman-MF.tif); %读入原图像文件imshow(I); %显示原图像fftI=fft2(I); %二维离散傅立叶变换sfftI=fftshift(fftI); %直流分量移到频谱中心RR=real(sfftI); %取傅立叶变换的实部II=imag(sfftI); %取傅立叶变换的虚部A=sqrt(RR.2+II.2); %计算频谱幅值A=(A-min(min(A)/(max(max(A)-min(min(A)*225; %归一化figure ;imshow(A); %显示失真图像傅立叶变化频谱图MF,map=imread(pout-MF.tif);imshow(MF);INITPSF=fspecial(motion,LEN,THETA); %将从获得失真参数的函数中获取的长度、角度失%参数带入进行运动失真图像的还原J,P=deconvblind(MF,INITPSF,30); %去卷积，恢复失真图像figure(3);imshow(J); function resim = Lucy(ifbl, LEN, THETA, iterations, handle)%Function to restore the image using Lucy-Richardson%Inputs: ifbl, LEN, THETA, iterations.%Returns: resim%ifbl: It is the input image.%THETA: It is the blur angle. The angle at which the image is blurred.%LEN: It is the blur length. The length is the number% of pixels by which the image is blurred.%iterations: It is the number of iterations.%handle:It is the handle to the waitbar(progress bar).%resim: It is the restored image.%Example:% resim = Lucy(image, LEN, THETA, iterations);% This call takes image, blur length, blur angle & no. of iterations % as input and returns the restored imageifbl = medfilt2(abs(ifbl);%Preprocessing%Performing Median Filter before restoring the blurred imageest = ifbl;%Initialising the initial estimate to the blurred imagePSF = fspecial(motion,LEN,THETA); %Create PSF of degradationOTF = psf2otf(PSF,size(ifbl); %Convert psf to otf of desired size%OTF is Optical Transfer Functioni = 1;while i=iterations fest = fft2(est); %Converting the estimate to frequency domain fblest = OTF.*fest; %Multiplying OTF with the estimate in frequency domain ifblest = ifft2(fblest); %Converting the blurred image estimate to spatial domain iratio = ifbl./ifblest; %Calculating ratio of blurred image and estimate of the deblurred %image firatio = fft2(iratio); %Converting the ratio to frequency domain corrvec = OTF .* firatio; %Calculating the correction vector icorrvec = ifft2(corrvec); %Converting the correction vector to spatial domain（ aftercorr = icorrvec.*est; %Multiplying correction vector & estimate of deblurred image to %find next estimate est = aftercorr; waitbar(i/iterations, handle); %Setting the waitbar indicating how much is completed i = i+1;endresim = abs(est); %Restored image四、测试和调试C=imread(cameraman.tif); subplot(1,2,1);imshow(C); LEN=35;THETA=50;PSF=fspecial(motion,LEN,THETA); MF=imfilter(C,PSF,circular,conv); subplot(1,2,2),imshow(MF);imwrite(MF,cameraman-MF.tif); I=imread(cameraman-MF.tif); imshow(I); fftI=fft2(I); sfftI=fftshift(fftI); RR=real(sfftI); II=imag(sfftI); A=sqrt(RR.2+II.2); A=(A-min(min(A)/(max(max(A)-min(min(A)*225;figure ;imshow(A); MF,map=imread(pout-MF.tif);imshow(MF);INITPSF=fspecial(motion,LEN,THETA);J,P=deconvblind(MF,INITPSF,30); figure(3);imshow(J); function resim = Lucy(ifbl, LEN, THETA, iterations, handle)ifbl