特色栏目： python 批处理 net编程 Javascript Php Asp Css Html5 Android seo centos

【图像去噪】基于二维双边高斯滤波实现图像去噪附matlab代码

来源：互联网收集：自由互联发布时间：2022-06-18

1 简介图像是生活中重要的信息来源,处理图像有助于理解信息的基本信息.但图像本身可能存在一些被干扰的信息或者噪声.研究了基于高斯滤波和双边滤波算法的数字图像处理技术用于

1 简介

图像是生活中重要的信息来源,处理图像有助于理解信息的基本信息.但图像本身可能存在一些被干扰的信息或者噪声.研究了基于高斯滤波和双边滤波算法的数字图像处理技术用于对图像的噪声进行消除.通过对图像进行理论模拟和实际图像处理,对两种算法进行比较和模拟得出相应结论.为今后的噪声消除方法的选择和实际工作的改进提供数据参考和依据.

2 完整代码

%Demo
%二维双边高斯滤波
close all
clear all
clc
%Bilateral filter
I1=imread('image1.jpg');
I2=I1; %use the same image for traditional bilateral filter
result=bfilter2(I1,I2,15,2,0.15);
figure;
imshow(result);
title('Bilateral filter');
imwrite(result,'result1.jpg');
%Joint-bilateral filter
I2=imread('image2.jpg');
result=bfilter2(I1,I2,15,2,0.15);
figure;
imshow(result);
title('Joint-bilateral filter');
imwrite(result,'result2.jpg');function img_out = bfilter2(image1, image2, n, sigma1, sigma2)
%bfilter2 function: perfrom two dimensional bilateral gaussian filtering.
%The standard deviations of the bilateral filter are given by sigma1 and
%sigma2, where the standard deviation of spatial-domain is given by sigma1
% and the standard deviation intensity-domain is given by sigma2.
%This function presents both bilateral filter and joint-bilateral filter.
%If you use the same image as image1 and image2, it is the normal bilateral
%filter; however, if you use different images in image1 and image2, you can
%use it as joint-bilateral filter, where the intensity-domain (range weight)
%calculations are performed using image2 and the spatial-domain (space weight)
%calculations are performed using image1.
%Usage:
% %Example1: normal bilateral filter using 5x5 kernel, spatial-sigma=6, and
% %intensity-sigma= 0.25:
% image=bfilter2(I1,I1,5,1.2,0.25);
% %Example2: joint-bilateral filter using 5x5 kernel, spatial-sigma=1.2,
% %and range-sigma= 0.25, the spatial-domain calculations are performed
% %using image (I1) and the intensity-domain calulcations (range weight)
% %are performed using image (I2):
% image=bfilter2(I1,I2,5,1.2,0.25);
% %Example3: use the default values for n, sigma1, and sigma2
% image=bfilter2(I1);
%Input:
% -image1: the spatial-domain image
% -image2: the intensity-domain (range weight) image (use the same image
% for the normal bilateral filter. Use different images for joint-bilateral
% filter.
% (default, use the same image; i.e. image2=image1)
% -n: kernel (window) size [nxn], should be odd number (default=5)
% -sigma1: the standard deviation of spatial-domain (default=1.2)
% sigma2: the standard deviation of intensity-domain (default=0.25)
%Author: Mahmoud Afifi, York University.
%argument's check
if nargin<1
error('Too few input arguments');
elseif nargin<2
image2=image1;
n=5;
sigma1=1.2;
sigma2=0.25;
elseif nargin<3
n=5;
sigma1=1.2;
sigma2=0.25;
elseif nargin<4
sigma1=1.2;
sigma2=0.25;
elseif nargin<5
sigma2=0.25;
end
%kernel size check
if mod(n,2)==0
error('Please use odd number for kernel size');
end
%dimensionality check
if size(image1,1)~=size(image2,1) || size(image1,2)~=size(image2,2) || ...
size(image1,3)~=size(image2,3)
error('Both images should have the same dimensions and number of color channels');
end
display('processing...');
w=floor(n/2);
% spatial-domain weights.
[X,Y] = meshgrid(-w:w,-w:w);
gs = exp(-(X.^2+Y.^2)/(2*sigma1^2));
%normalize images
if isa(image1,'uint8')==1
image1=double(image1)/255;
end
if isa(image2,'uint8')==1
image2=double(image2)/255;
end
%intialize img_out
img_out=zeros(size(image1,1),size(image1,2),size(image1,3));
%padd both iamges
image1=padarray(image1,[w w],'replicate','both');
image2=padarray(image2,[w w],'replicate','both');
for i=ceil(n/2):size(image1,1)-w
for j=ceil(n/2):size(image1,2)-w
patch1(:,:,:)=image1(i-w:i+w,j-w:j+w,:);
patch2(:,:,:)=image2(i-w:i+w,j-w:j+w,:);
d=(repmat(image2(i,j,:),[n,n])-patch2).^2;
% intensity-domain weights. (range weights)
gr=exp(-(d)/(2*sigma2^2));
for c=1:size(image1,3)
g(:,:,c)=gs.*gr(:,:,c); %bilateral filter
normfactor=1/sum(sum(g(:,:,c))); %normalization factor
%apply equation:
%out[i]=normfactor*sum (kernel * image)
img_out(i-ceil(n/2)+1,j-ceil(n/2)+1,c)=...
sum(sum(g(:,:,c).*patch1(:,:,c)))*normfactor;
% imshow(img_out,[]);
end
end
end
img_out=uint8(img_out*255);
end