数字图像处理小波变换.ppt
- 格式:ppt
- 大小:2.29 MB
- 文档页数:67
第1章Haar小波分析1.1简介(近距离---小尺度) (高分辨率)(远距离---大尺度) (低分辨率)1.2 平均与细节设1234{,,,}x x x x 是一个信号序列。
定义它的平均和细节:1,0121,012()/2()/2a x x d x x =+⎫⎬=-⎭找出了1x 、2x 和1,0a 、1,0d 的关系。
这里,1,0a 是原信号前两个值1x 、2x 的平均。
又叫低频成分,反映前两个值1x 、2x 的基本特征或粗糙趋势;1,0d 反映了1x 、2x 的差别,即细节信息,又叫高频成分。
1,1341,134()/2()/2a x x d x x =+⎫⎬=-⎭找出了3x 、4x 和1,1a 、1,1d 的关系。
同样,1,1a 是原信号后两个值3x 、4x 的平均,1,1d 反映了3x 、4x 的细节。
我们把1,01,11,01,1{,,,}a a d d 看作是对1234{,,,}x x x x 实施了一次变换的结果。
变换还可以往下进行:0,01,01,1()/2a a a =+=1234(()/2()/2)/2x x x x +++ =1234()/4x x x x +++0,0a 是对4个信号元素最终的平均,它是原信号最基本的信息;0,01,01,1()/2d a a =-。
经过二次变换,我们得到了原信号的另一种表示:0,00,01,01,1{,,,}a d d d该序列叫做原序列的小波变换,0,00,01,01,1,,,a d d d 叫做小波系数。
还可以反过来表示:111,0211,0x a d x a d =+⎫⎬=-⎭这是用{1a ,1,0d }来恢复原信号1x 、2x ;321,1421,1x a d x a d =+⎫⎬=-⎭用{2a ,1,1d }来恢复原信号3x 、4x 。
也就是反变换。
小波变换过程的塔式算法:例如,1234{,,,}x x x x ={3,1,-2,4}最终的小波变换为0,00,01,01,1{,,,}a d d d =31{,,1,3}22-1.3 尺度函数与小波函数 (1)Haar 尺度函数不压缩:不位移 位移一个单位 位移k 个单位t1)-压缩1/12倍,不位移压缩1/12倍,位移一个单位 压缩1/2j倍,移位K 个单位一般,()(2)j j k t t k φφ=-,0,1,2,...,21j k =-◆ 几个术语1) 支撑(支集),(尺度)函数,()j k t φ不为零的区间,上例中为1[,]22j j k k +。
(1) 名词解释RGB Red Green Blue,红绿蓝三原色CMYK Cyan Magenta yellow blacK , 用于印刷的四分色HIS Horizontal Situation Indicator 水平位置指示器FFT Fast Fourier Transform Algorithm (method) 快速傅氏变换算法CWT continuous wavelet transform 连续小波变换DCT Discrete Cosine Transform 离散余弦变换DWT DiscreteWaveletTransform 离散小波变换CCD Charge Coupled Device 电荷耦合装置Pixel: a digital image is composed of a finite number of elements,each of which has a particular lication and value,these elements are called pixel 像素DC component in frequency domain 频域直流分量GLH Gray Level Histogram 灰度直方图Mather(basic)wavelet:a function (wave) used to generate a set of wavelets, 母小波,用于产生小波变换所需的一序列子小波Basis functions basis image 基函数基图像Multi-scale analysis 多尺度分析Gaussian function 高斯函数sharpening filter 锐化滤波器Smoothing filter/convolution 平滑滤波器/卷积Image enhancement /image restoration 图像增强和图像恢复(2)问答题1. Cite one example of digital image processingAnswer: In the domain of medical image processing we may need to inspect a certain class of images generated by an electron microscope to eliminate bright, isolated dots that are no interest.2.Cite one example of frequency domain operation from the following processing result, make a general comment about ideal highpass filter (figure B) and Gaussian highpass filter(figure D)A. Original imageB. ideal highpass filterIn contrast to the ideal low pass filter, it is to let all the signals above the cutoff frequency fc without loss, and to make all the signals below the cutoff frequency of FC without loss of.C. the result of ideal highpass filterD. Gaussian highpass filterHigh pass filter, also known as "low resistance filter", it is an inhibitory spectrum of the low frequency signal and retain high frequency signal model (or device). High pass filter can make the high frequency components, while the high-frequency part of the frequency in the image of the sharp change in the gray area, which is often the edge of the object. So high pass filter can make the image get sharpening processingE. The result of Gaussian filter3.The original image, the ideal lowpass filter and Gaussian lowpass filter are shown below B nd C .D and E are the result of the eitherfilter B or CA. Draw lines to connect the filter with their resultB. Explain the difference of the two filtersDue to excessive characteristics of the ideal low-pass filter too fast Jun, it will produce a ringing phenomenon.Over characteristics of Gauss filter is very flat, so it is not ringing4.What is the result when applying an averaging mask with the size 1X1?5.State the concept of the Nyquist sampling theorem from the figure belovyThe law of sampling process should be followed, also called the sampling theorem and the sampling theorem. The sampling theorem showsthe relationship between the sampling frequency and the signal spectrum, and it is the basic basis of the continuous signal discretization. In analog / digital signal conversion process, when the sampling frequency fs.max greater than 2 times the highest frequency present in the signal Fmax fs.max>2fmax, sampling digital signal completely retained the information in the original signal, the general practical application assurance sampling frequency is 5 ~ 10 times higher than that of the signal of the high frequency; sampling theorem, also known as the Nyquist theorem6.A mean filter is a linear filter but a median filter is not, why?Mean filter is a typical linear filtering algorithm, it is to point to in the target pixels in the image to a template, this template including its surrounding adjacent pixels and the pixels in itself.To use in the template to replace all the pixels of average pixelvalues.Linear filter, median filter, also known as the main method used in the bounded domain average method.Median filter is a kind of commonly used nonlinear smoothing filter and its basic principle is to put the little value in a digital image or sequence to use value at various points in the field of a point at which the value to replace, its main function is to let the surrounding pixel gray value differences between larger pixel change with the surrounding pixels value close to the values, which can eliminate the noise of the isolated points, so median filter to filter out the salt and pepper noise image is very effective.(3)算法题1.The following matrix A is a 3*3 image and B is 3*3 Laplacian mask, what will be the resulting image? (Note that the elements beyond the border remain unchanged)2.Develop an algorithm to obtain the processing result B from original image A3.Develop an algorithm which computes the pseudocolor image processing by means of fourier tramsformAnswer:The steps of the process are as follow:(1) Multiply the input image f(x,y) by (-1)x+y tocenter the transform;(2) Compute the DFT of the image from (1) to get power spectrumF(u,v) of Fourier transform.(3) Multiply by a filter function h(u,v) .(4) Compute the inverse DFT of the result in (3).(5) Obtain the real part of the result in (4).(6) Multiply the result in (5) by(-1)x+y4.Develop an algorithm to generate approximation image series shown in the following figure b** means of down sampling.(4)编程题There are two satellite photos of night as blew.Write a programwith MATLAB to tell which is brighterAn 8*8 image f(i,i) has gray levels given by the following equation:f(i,i)=|i-j|, i,j=0,1 (7)Write a program to find the output image obtained by applying a 3*3 median filter on the image f(i,j) ;note that the border pixels remain unchanged.Answer:1.Design an adaptive local noise reduction filter and apply it to an image with Gaussian noise. Compare the performance of the adaptive local noise reduction filter with arithmetic mean and geometric mean filter.Answer:clearclose all;rt=imread('E:\数字图像处理\yy.bmp');gray=rgb2gray(rt);subplot(2,3,1);imshow(rt);title('原图像') ;subplot(2,3,2);imshow(gray);title('原灰度图像') ;rtg=im2double(gray);rtg=imnoise(rtg,'gaussian',0,0.005)%加入均值为0,方差为0.005的高斯噪声subplot(2,3,3);imshow(rtg);title('高噪点处理后的图像');[a,b]=size(rtg);n=3;smax=7;nrt=zeros(a+(smax-1),b+(smax-1));for i=((smax-1)/2+1):(a+(smax-1)/2)for j=((smax-1)/2+1):(b+(smax-1)/2)nrt(i,j)=rtg(i-(smax-1)/2,j-(smax-1)/2);endendfigure;imshow(nrt);title('扩充后的图像');nrt2=zeros(a,b);for i=n+1:a+nfor j=n+1:b+nfor m1=3:2m2=(m1-1)/2;c=nrt2(i-m2:i+m2,j-m2:j+m2);%使用7*7的滤波器Zmed=median(median(c));Zmin=min(min(c));Zmax=max(max(c));A1=Zmed-Zmin;A2=Zmed-Zmax;if(A1>0&&A2<0)B1=nrt2(i,j)-Zmin;B2=nrt2(i,j)-Zmax;if(B1>0&&B2<0)nrt2(i,j)= nrt2(i,j);elsenrt2(i,j)=Zmed;endcontinue;endendendendnrt3=im2uint8(nrt2);figure;imshow(nrt3);title('自适应中值滤波图');2. Implement Wiener filter with “wiener2” function of MatLab to an image with Gaussian noise and compare the performance with adaptive local noise reduction filter.代码如下:>> I=imread('E:\数字图像处理\yy.bmp');>>J=rgb2gray(I);>>K = imnoise(J,'gaussian',0,0.005);>>L=wiener2(K,[5 5]);>>subplot(1,2,1);imshow(K);title('高噪点处理后的图像');>>subplot(1,2,2);imshow(L);title('维纳滤波器处理后的图像');3. Image smoothing with arithmetic averaging filter (spatial convolution).图像平滑与算术平均滤波(空间卷积)。
小波变换在图像处理中的应用及其实例引言:随着数字图像处理技术的不断发展,小波变换作为一种重要的数学工具,被广泛应用于图像处理领域。
小波变换具有多尺度分析的特点,能够提取图像的局部特征,对图像进行有效的压缩和去噪处理。
本文将探讨小波变换在图像处理中的应用,并通过实例加以说明。
一、小波变换的基本原理小波变换是将信号或图像分解成一组基函数,这些基函数是由母小波函数进行平移和伸缩得到的。
小波变换的基本原理是将信号或图像在不同尺度上进行分解,得到不同频率的小波系数,从而实现信号或图像的分析和处理。
二、小波变换在图像压缩中的应用图像压缩是图像处理中的重要应用之一。
小波变换通过分解图像,将图像的高频和低频信息分离出来,从而实现图像的有损或无损压缩。
小波变换在图像压缩中的应用主要有以下两个方面:1. 小波变换在JPEG2000中的应用JPEG2000是一种新一代的图像压缩标准,它采用小波变换作为核心算法。
JPEG2000通过小波变换将图像分解成多个子带,然后对每个子带进行独立的压缩,从而实现对图像的高效压缩。
相比于传统的JPEG压缩算法,JPEG2000在保持图像质量的同时,能够更好地处理图像的细节和边缘信息。
2. 小波变换在图像去噪中的应用图像去噪是图像处理中的常见问题,而小波变换能够有效地去除图像中的噪声。
小波变换通过将图像分解成多个尺度的小波系数,对每个尺度的小波系数进行阈值处理,将较小的小波系数置零,从而抑制图像中的噪声。
经过小波变换去噪后的图像能够更清晰地显示图像的细节和边缘。
三、小波变换在图像增强中的应用图像增强是改善图像质量的一种方法,而小波变换能够提取图像的局部特征,从而实现图像的增强。
小波变换在图像增强中的应用主要有以下两个方面:1. 小波变换在图像锐化中的应用图像锐化是增强图像边缘和细节的一种方法,而小波变换能够提取图像的边缘信息。
通过对图像进行小波变换,可以得到图像的高频小波系数,然后对高频小波系数进行增强处理,从而增强图像的边缘和细节。