opencv 实现导向滤波

from：http://blog.csdn.net/wds555/article/details/23176313

何凯明去雾算法中的导向滤波实现，原文地址导向滤波。

导向图像I，滤波输入图像p以及输出图像q。像素点 i 处的滤波结果是被表达成一个加权平均：

假设导向滤波器在导向图像I和滤波输出q之间是一个局部线性模型：

最小化下面的窗口Wk的代价函数：

用来确定a,b的值

其中

论文所给算法如下：

matlab代码如下：

[plain] view plaincopy

function q = guidedfilter(I, p, r, eps)
% GUIDEDFILTER O(1) time implementation of guided filter.
%
% - guidance image: I (should be a gray-scale/single channel image)
% - filtering input image: p (should be a gray-scale/single channel image)
% - local window radius: r
% - regularization parameter: eps
[hei, wid] = size(I);
N = boxfilter(ones(hei, wid), r); % the size of each local patch; N=(2r+1)^2 except for boundary pixels.
mean_I = boxfilter(I, r) ./ N;
mean_p = boxfilter(p, r) ./ N;
mean_Ip = boxfilter(I.*p, r) ./ N;
cov_Ip = mean_Ip - mean_I .* mean_p; % this is the covariance of (I, p) in each local patch.
mean_II = boxfilter(I.*I, r) ./ N;
var_I = mean_II - mean_I .* mean_I;
a = cov_Ip ./ (var_I + eps); % Eqn. (5) in the paper;
b = mean_p - a .* mean_I; % Eqn. (6) in the paper;
mean_a = boxfilter(a, r) ./ N;
mean_b = boxfilter(b, r) ./ N;
q = mean_a .* I + mean_b; % Eqn. (8) in the paper;
end

[plain] view plaincopy

function imDst = boxfilter(imSrc, r)
% BOXFILTER O(1) time box filtering using cumulative sum
%
% - Definition imDst(x, y)=sum(sum(imSrc(x-r:x+r,y-r:y+r)));
% - Running time independent of r;
% - Equivalent to the function: colfilt(imSrc, [2*r+1, 2*r+1], 'sliding', @sum);
% - But much faster.
[hei, wid] = size(imSrc);
imDst = zeros(size(imSrc));
%cumulative sum over Y axis
imCum = cumsum(imSrc, 1);
%difference over Y axis
imDst(1:r+1, :) = imCum(1+r:2*r+1, :);
imDst(r+2:hei-r, :) = imCum(2*r+2:hei, :) - imCum(1:hei-2*r-1, :);
imDst(hei-r+1:hei, :) = repmat(imCum(hei, :), [r, 1]) - imCum(hei-2*r:hei-r-1, :);
%cumulative sum over X axis
imCum = cumsum(imDst, 2);
%difference over Y axis
imDst(:, 1:r+1) = imCum(:, 1+r:2*r+1);
imDst(:, r+2:wid-r) = imCum(:, 2*r+2:wid) - imCum(:, 1:wid-2*r-1);
imDst(:, wid-r+1:wid) = repmat(imCum(:, wid), [1, r]) - imCum(:, wid-2*r:wid-r-1);
end

以下为单通道图像导向滤波opencv实现：

[cpp] view plaincopy

#include "myGuidedFilter_Mat.h"
CvMat * cumsum(CvMat *src,int rc)
{
CvMat *Imdst = cvCreateMat(src->rows,src->cols,CV_64FC1);
Imdst=cvCloneMat(src);
if (rc==1)
{
for(int y=1;y<src->height;y++)
{
double *ptr0=(double *)(Imdst->data.ptr+(y-1)*Imdst->step);
double *ptr=(double *)(Imdst->data.ptr+y*Imdst->step);
for(int x=0;x<src->width;x++)
{
ptr[x]=ptr0[x]+ptr[x];
//cvSetReal2D(Imdst,y,x,cvGetReal2D(Imdst,y-1,x)+cvGetReal2D(Imdst,y,x));
}
}
}
else if (rc==2)
{
for(int y=0;y<src->height;y++)
{
double *ptr=(double *)(Imdst->data.ptr+y*Imdst->step);
for(int x=1;x<src->width;x++)
{
ptr[x]=ptr[x-1]+ptr[x];
//cvSetReal2D(Imdst,y,x,cvGetReal2D(Imdst,y,x-1)+cvGetReal2D(Imdst,y,x));
}
}
}
return Imdst;
}
CvMat * boxFilter(CvMat *src,int r)
{
CvMat *Imdst = cvCreateMat(src->rows,src->cols,CV_64FC1);
Imdst=cvCloneMat(src);
CvMat *subImage;
//imCum = cumsum(imSrc, 1);
CvMat *imCum = cumsum(Imdst,1);
//imDst(1:r+1, :) = imCum(1+r:2*r+1, :);
for (int y = 0;y<r;y++)
{
//double *ptrDst=(double *)Imdst->data.ptr+y*Imdst->step;
//double *ptrCum=(double *)imCum->data.ptr+(y+r)*imCum->step;
for(int x = 0;x<Imdst->width;x++)
{
//ptrDst[x]=ptrCum[x];
cvSetReal2D(Imdst,y,x,cvGetReal2D(imCum,y+r,x));
}
}
//imDst(r+2:hei-r, :) = imCum(2*r+2:hei, :) - imCum(1:hei-2*r-1, :);
for (int y = r+1;y<Imdst->height-r-1;y++)
{
for(int x = 0;x<Imdst->width;x++)
{
cvSetReal2D(Imdst,y,x,(cvGetReal2D(imCum,y+r,x)-cvGetReal2D(imCum,y-r-1,x)));
}
}
//imDst(hei-r+1:hei, :) = repmat(imCum(hei, :), [r, 1]) - imCum(hei-2*r:hei-r-1, :);
subImage = cvCreateMat(r,Imdst->width,CV_64FC1);
CvMat *tem=cvCreateMat(1,Imdst->width,CV_64FC1);
cvGetRow(imCum,tem,imCum->height-1);
cvRepeat(tem,subImage);
/*for(int y=0;y<r;y++)
{
for(int x=0;x<Imdst->width;x++)
{
cvSetReal2D(subImage,y,x,cvGetReal2D(imCum,Imdst->height-1,x));
}
}*/
for (int y = Imdst->height-r;y<Imdst->height;y++)
{
for(int x = 0;x<Imdst->width;x++)
{
cvSetReal2D(Imdst,y,x,cvGetReal2D(subImage,y-Imdst->height+r,x)-cvGetReal2D(imCum,y-r-1,x));
}
}
cvReleaseMat(&subImage);
cvReleaseMat(&tem);
imCum = cumsum(Imdst, 2);
//imDst(:, 1:r+1) = imCum(:, 1+r:2*r+1);
for (int y = 0;y<Imdst->height;y++)
{
for(int x = 0;x<r;x++)
{
cvSetReal2D(Imdst,y,x,cvGetReal2D(imCum,y,x+r));
}
}
//imDst(:, r+2:wid-r) = imCum(:, 2*r+2:wid) - imCum(:, 1:wid-2*r-1);
for (int y = 0;y<Imdst->height;y++)
{
for(int x = r+1;x<Imdst->width-r-1;x++)
{
cvSetReal2D(Imdst,y,x,(cvGetReal2D(imCum,y,x+r)-cvGetReal2D(imCum,y,x-r-1)));
}
}
//imDst(:, wid-r+1:wid) = repmat(imCum(:, wid), [1, r]) - imCum(:, wid-2*r:wid-r-1);
subImage = cvCreateMat(Imdst->height,r,CV_64FC1);
tem=cvCreateMat(Imdst->height,1,CV_64FC1);
cvGetCol(imCum,tem,imCum->width-1);
cvRepeat(tem,subImage);
/*for(int y=0;y<Imdst->height;y++)
{
for(int x=0;x<r;x++)
{
cvSetReal2D(subImage,y,x,cvGetReal2D(imCum,y,Imdst->width-1));
}
}*/
for (int y = 0;y<Imdst->height;y++)
{
for(int x = Imdst->width-r;x<Imdst->width;x++)
{
cvSetReal2D(Imdst,y,x,cvGetReal2D(subImage,y,x-Imdst->width+r)-cvGetReal2D(imCum,y,x-r-1));
}
}
cvReleaseMat(&subImage);
return Imdst;
}
CvMat * myGuidedFilter_Mat(CvMat * I,CvMat *img_pp,int r, double eps)
{
int height = img_pp->height;
int width = img_pp->width;
int type = CV_64FC1;
CvMat *ones = cvCreateMat(height,width,type);
cvSet(ones,cvRealScalar(1));
CvMat * N = boxFilter(ones,r);
//求I的均值
CvMat * mean_I = cvCreateMat(height,width,type);
cvDiv(boxFilter(I,r),N,mean_I);
//求P的均值
CvMat * mean_p = cvCreateMat(height,width,type);
cvDiv(boxFilter(img_pp,r),N,mean_p);
//求I*P的均值
CvMat * pr = cvCreateMat(height,width,type);
cvMul(I,img_pp,pr);
CvMat * mean_Ip = cvCreateMat(height,width,type);
cvDiv(boxFilter(pr,r),N,mean_Ip);
//求I与p协方差
cvMul(mean_I,mean_p,pr);
CvMat * cov_Ip = cvCreateMat(height,width,type);
cvSub(mean_Ip,pr,cov_Ip);
//求I的方差
CvMat * var_I = cvCreateMat(height,width,type);
cvMul(I,I,pr);
cvDiv(boxFilter(pr,r),N,var_I);
cvMul(mean_I,mean_I,pr);
cvSub(var_I,pr,var_I);
//求a
CvMat * a = cvCreateMat(height,width,type);
cvAddS(var_I,cvScalar(eps),var_I);
cvDiv(cov_Ip,var_I,a);
//求b
CvMat * b = cvCreateMat(height,width,type);
cvMul(a,mean_I,pr);
cvSub(mean_p,pr,b);
//求a的均值
CvMat * mean_a = cvCreateMat(height,width,type);
cvDiv(boxFilter(a,r),N,mean_a);
//求b的均值
CvMat * mean_b = cvCreateMat(height,width,type);
cvDiv(boxFilter(b,r),N,mean_b);
//求Q
CvMat * q = cvCreateMat(height,width,type);
cvMul(mean_a,I,a);
cvAdd(a,mean_b,q);
cvReleaseMat(&ones);
cvReleaseMat(&mean_I);
cvReleaseMat(&mean_p);
cvReleaseMat(&pr);
cvReleaseMat(&mean_Ip);
cvReleaseMat(&cov_Ip);
cvReleaseMat(&var_I);
cvReleaseMat(&a);
cvReleaseMat(&b);
cvReleaseMat(&mean_a);
cvReleaseMat(&mean_b);
return q;
}

opencv 实现导向滤波相关推荐

OpenCV中导向滤波介绍与应用
点击上方"小白学视觉",选择加"星标"或"置顶" 重磅干货,第一时间送达导向滤波介绍导向滤波是使用导向图像作为滤波内容图像,在导向图像上 ...
详解——导向滤波(Guided Filter)和快速导向滤波
文章目录导读原理推导导向滤波的应用导向滤波的实现快速导向滤波的实现算法效果代码参考导读在图像滤波算法中,导向滤波.双边滤波.最小二乘滤波并称三大保边滤波器,他们是各向异性滤波器.相 ...
【OpenCV 例程200篇】61. 导向滤波（Guided filter）
[OpenCV 例程200篇]61. 导向滤波(Guided filter) 欢迎关注『OpenCV 例程200篇』系列,持续更新中欢迎关注『Python小白的OpenCV学习课』系列,持续 ...
快速导向滤波 matlab,导向滤波小结：从导向滤波（guided filter）到快速导向滤波（fast guide filter）的原理，应用及opencv实现代码...
http://blog.csdn.net/kuweicai/article/details/78385871 1. 导向滤波简介导向滤波是何凯明在学生时代提出的一个保边滤波(edge-preserv ...
【OpenCV图像处理】十七、图像的导向滤波
http://blog.csdn.net/qq_34784753/article/details/70229009?locationNum=12&fps=1 导向图滤波是一种图像滤波技术,通过 ...
导向滤波小结：从导向滤波（guided filter）到快速导向滤波（fast guide filter）的原理，应用及opencv实现代码
1. 导向滤波简介导向滤波是何凯明在学生时代提出的一个保边滤波(edge-preserving smoothing)算法.何凯明在cv圈应该算是名人了,学生时代关于图像去雾的研究就以第一作者的身份获 ...
在OpenCV里实现导向滤波
前面学习了双边滤波和联合滤波,都可以保边滤波,计算方式比较类似,都是使用相似性权重模板,下面来介绍一种不依赖于权重模板的保边滤波的另外一种方法-导向滤波.导向滤波比前面两种滤波优点有计算速度快,并且细 ...
Opencv-Python-导向滤波快速导向滤波
版本:Python:2.7.15 OpenCV:2.4.13 导向滤波算法原理原理可以看博主:白马负金羁的文章导向滤波(Guided Filter)的解析与实现,对原理解释十分通俗易懂. 导向滤 ...
导向滤波快速导向滤波及引导图的选择
引导图的选择我主要说的导向滤波其中的引导图选择问题. 百度百科的定义 : 导向图滤波是一种图像滤波技术 ,通过一张引导图G(导向图),对目标图像P(输入图像)进行滤波处理,使得最后的输出图像大体上与 ...

opencv 实现导向滤波

opencv 实现导向滤波相关推荐

最新文章

热门文章