双边滤波算法原理:
转载自:https://blog.csdn.net/piaoxuezhong/article/details/78302920
双边滤波是一种非线性滤波器,它可以达到保持边缘、降噪平滑的效果。和其他滤波原理一样,双边滤波也是采用加权平均的方法,用周边像素亮度值的加权平均代表某个像素的强度,所用的加权平均基于高斯分布[1]。最重要的是,双边滤波的权重不仅考虑了像素的欧氏距离(如普通的高斯低通滤波,只考虑了位置对中心像素的影响),还考虑了像素范围域中的辐射差异(例如卷积核中像素与中心像素之间相似程度、颜色强度,深度距离等),在计算中心像素的时候同时考虑这两个权重。
公式1a,1b给出了双边滤过的操作,Iq为输入图像,Ipbf为滤波后图像:
mark下双边滤波里的两个权重域的概念:空间域(spatial domain S)和像素范围域(range domain R),这个是它跟高斯滤波等方法的最大不同点。下面是我找到的对比说明,更好地理解双边滤波,首先是高斯滤波的情况:
然后对比再看一下双边滤波的过程:
双边滤波的核函数是空间域核与像素范围域核的综合结果:在图像的平坦区域,像素值变化很小,对应的像素范围域权重接近于1,此时空间域权重起主要作用,相当于进行高斯模糊;在图像的边缘区域,像素值变化很大,像素范围域权重变大,从而保持了边缘的信息。
为了更加形象的说明两个权重的影响,作者还给出了二维图像的直观说明:
双边滤波算法C++实现
#include <opencv2/core.hpp>
#include <opencv2/highgui.hpp>
#include <opencv2/imgproc.hpp>
#include <opencv2/highgui/highgui_c.h>
#include <iostream>
////////////////////////////
//获取高斯模板(空间模板)
///////////////////////////
void getGausssianMask(cv::Mat& Mask, cv::Size wsize, double spaceSigma) {
Mask.create(wsize, CV_64F);
int h = wsize.height;
int w = wsize.width;
int center_h = (h - 1) / 2;
int center_w = (w - 1) / 2;
double sum = 0.0;
double x, y;
for (int i = 0; i < h; ++i) {
y = pow(i - center_h, 2);
double* Maskdate = Mask.ptr<double>(i);
for (int j = 0; j < w; ++j) {
x = pow(j - center_w, 2);
double g = exp(-(x + y) / (2 * spaceSigma * spaceSigma));
Maskdate[j] = g;
sum += g;
}
}
}
////////////////////////////
//获取色彩模板(值域模板)
///////////////////////////
void getColorMask(std::vector<double> &colorMask, double colorSigma) {
for (int i = 0; i < 256; ++i) {
double colordiff = exp(-(i*i) / (2 * colorSigma * colorSigma));
colorMask.push_back(colordiff);
}
}
////////////////////////////
//双边滤波
///////////////////////////
void bilateralfiter(cv::Mat& src, cv::Mat& dst, cv::Size wsize, double spaceSigma, double colorSigma) {
cv::Mat spaceMask;
std::vector<double> colorMask;
cv::Mat Mask0 = cv::Mat::zeros(wsize, CV_64F);
cv::Mat Mask1 = cv::Mat::zeros(wsize, CV_64F);
cv::Mat Mask2 = cv::Mat::zeros(wsize, CV_64F);
getGausssianMask(spaceMask, wsize, spaceSigma);//空间模板
getColorMask(colorMask, colorSigma);//值域模板
int hh = (wsize.height - 1) / 2;
int ww = (wsize.width - 1) / 2;
dst.create(src.size(), src.type());
//边界填充
cv::Mat Newsrc;
cv::copyMakeBorder(src, Newsrc, hh, hh, ww, ww, cv::BORDER_REPLICATE);//边界复制;
for (int i = hh; i < src.rows + hh; ++i) {
for (int j = ww; j < src.cols + ww; ++j) {
double sum[3] = { 0 };
int graydiff[3] = { 0 };
double space_color_sum[3] = { 0.0 };
for (int r = -hh; r <= hh; ++r) {
for (int c = -ww; c <= ww; ++c) {
if (src.channels() == 1) {
int centerPix = Newsrc.at<uchar>(i, j);
int pix = Newsrc.at<uchar>(i + r, j + c);
graydiff[0] = abs(pix - centerPix);
double colorWeight = colorMask[graydiff[0]];
Mask0.at<double>(r + hh, c + ww) = colorWeight * spaceMask.at<double>(r + hh, c + ww);//滤波模板
space_color_sum[0] = space_color_sum[0] + Mask0.at<double>(r + hh, c + ww);
}
else if (src.channels() == 3) {
cv::Vec3b centerPix = Newsrc.at<cv::Vec3b>(i, j);
cv::Vec3b bgr = Newsrc.at<cv::Vec3b>(i + r, j + c);
graydiff[0] = abs(bgr[0] - centerPix[0]); graydiff[1] = abs(bgr[1] - centerPix[1]); graydiff[2] = abs(bgr[2] - centerPix[2]);
double colorWeight0 = colorMask[graydiff[0]];
double colorWeight1 = colorMask[graydiff[1]];
double colorWeight2 = colorMask[graydiff[2]];
Mask0.at<double>(r + hh, c + ww) = colorWeight0 * spaceMask.at<double>(r + hh, c + ww);//滤波模板
Mask1.at<double>(r + hh, c + ww) = colorWeight1 * spaceMask.at<double>(r + hh, c + ww);
Mask2.at<double>(r + hh, c + ww) = colorWeight2 * spaceMask.at<double>(r + hh, c + ww);
space_color_sum[0] = space_color_sum[0] + Mask0.at<double>(r + hh, c + ww);
space_color_sum[1] = space_color_sum[1] + Mask1.at<double>(r + hh, c + ww);
space_color_sum[2] = space_color_sum[2] + Mask2.at<double>(r + hh, c + ww);
}
}
}
//滤波模板归一化
if (src.channels() == 1)
Mask0 = Mask0 / space_color_sum[0];
else {
Mask0 = Mask0 / space_color_sum[0];
Mask1 = Mask1 / space_color_sum[1];
Mask2 = Mask2 / space_color_sum[2];
}
for (int r = -hh; r <= hh; ++r) {
for (int c = -ww; c <= ww; ++c) {
if (src.channels() == 1) {
sum[0] = sum[0] + Newsrc.at<uchar>(i + r, j + c) * Mask0.at<double>(r + hh, c + ww); //滤波
}
else if (src.channels() == 3) {
cv::Vec3b bgr = Newsrc.at<cv::Vec3b>(i + r, j + c); //滤波
sum[0] = sum[0] + bgr[0] * Mask0.at<double>(r + hh, c + ww);//B
sum[1] = sum[1] + bgr[1] * Mask1.at<double>(r + hh, c + ww);//G
sum[2] = sum[2] + bgr[2] * Mask2.at<double>(r + hh, c + ww);//R
}
}
}
for (int k = 0; k < src.channels(); ++k) {
if (sum[k] < 0)
sum[k] = 0;
else if (sum[k]>255)
sum[k] = 255;
}
if (src.channels() == 1)
{
dst.at<uchar>(i - hh, j - ww) = static_cast<uchar>(sum[0]);
}
else if (src.channels() == 3)
{
cv::Vec3b bgr = { static_cast<uchar>(sum[0]), static_cast<uchar>(sum[1]), static_cast<uchar>(sum[2]) };
dst.at<cv::Vec3b>(i - hh, j - ww) = bgr;
}
}
}
}
int main() {
cv::Mat src = cv::imread("D:\\z2.jpg");
if (src.empty()) {
return -1;
}
cv::Mat dst;
double spaceSigma = 10;//10
double colorSigma = 30;//30
cv::namedWindow("src", CV_WINDOW_NORMAL);
cv::imshow("src", src);
bilateralfiter(src, dst, cv::Size(23, 23), spaceSigma, colorSigma);
cv::namedWindow("双边滤波", CV_WINDOW_NORMAL);
cv::imshow("双边滤波", dst);
cv::imwrite("D:\\image\\z2.jpg", dst);
cv::waitKey(0);
return 0;
}
:
参考:
- https://en.wikipedia.org/wiki/Bilateral_filtering
- http://people.csail.mit.edu/sparis/bf_course/
- http://people.csail.mit.edu/sparis/bf/
- http://blog.csdn.net/fightingforcv/article/details/52723376
- http://blog.csdn.net/mumusan2016/article/details/54578038
- http://blog.csdn.net/majinlei121/article/details/50463514
- http://kaiminghe.com/eccv10/index.html (Guided Image Filtering)
- http://blog.csdn.net/dangchangying/article/details/14451963