Skip to content

Commit e7f614c

Browse files
BBufBBuf
committed
Update Canny Canny Edge Detection.cpp
1 parent 1ce9813 commit e7f614c

File tree

2 files changed

+181
-1
lines changed

2 files changed

+181
-1
lines changed

‎Feature Extraction/Canny Edge Detection.cpp

Lines changed: 178 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,178 @@
1+
#include "opencv2/opencv.hpp"
2+
#include "opencv2/imgproc/imgproc.hpp"
3+
#include "iostream"
4+
#include "algorithm"
5+
#include "vector"
6+
#include "stdio.h"
7+
using namespace std;
8+
using namespace cv;
9+
10+
const double PI = 3.1415926;
11+
12+
double getSum(Mat src){
13+
double sum = 0;
14+
for(int i = 0; i < src.rows; i++){
15+
for(int j = 0; j < src.cols; j++){
16+
sum += (double)src.at<double>(i, j);
17+
}
18+
}
19+
return sum;
20+
}
21+
22+
Mat CannyEdgeDetection(cv::Mat src, int kSize, double hightThres, double lowThres){
23+
// if(src.channels() == 3){
24+
// cvtColor(src, src, CV_BGR2GRAY);
25+
// }
26+
cv::Rect rect;
27+
Mat gaussResult;
28+
int row = src.rows;
29+
int col = src.cols;
30+
printf("%d %d\n", row, col);
31+
Mat filterImg = Mat::zeros(row, col, CV_64FC1);
32+
src.convertTo(src, CV_64FC1);
33+
Mat dst = Mat::zeros(row, col, CV_64FC1);
34+
int gaussCenter = kSize / 2;
35+
double sigma = 1;
36+
Mat guassKernel = Mat::zeros(kSize, kSize, CV_64FC1);
37+
for(int i = 0; i < kSize; i++){
38+
for(int j = 0; j < kSize; j++){
39+
guassKernel.at<double>(i, j) = (1.0 / (2.0 * PI * sigma * sigma)) * (double)exp(-(((double)pow((i - (gaussCenter+1)), 2) + (double)pow((j-(gaussCenter+1)), 2)) / (2.0*sigma*sigma)));
40+
}
41+
}
42+
Scalar sumValueScalar = cv::sum(guassKernel);
43+
double sum = sumValueScalar.val[0];
44+
guassKernel = guassKernel / sum;
45+
for(int i = gaussCenter; i < row - gaussCenter; i++){
46+
for(int j = gaussCenter; j < col - gaussCenter; j++){
47+
rect.x = j - gaussCenter;
48+
rect.y = i - gaussCenter;
49+
rect.width = kSize;
50+
rect.height = kSize;
51+
//printf("%d %d\n", i, j);
52+
//printf("%d %d %.5f\n", i, j, cv::sum(guassKernel.mul(src(rect))).val[0]);
53+
filterImg.at<double>(i, j) = cv::sum(guassKernel.mul(src(rect))).val[0];
54+
}
55+
}
56+
Mat gradX = Mat::zeros(row, col, CV_64FC1); //水平梯度
57+
Mat gradY = Mat::zeros(row, col, CV_64FC1); //垂直梯度
58+
Mat grad = Mat::zeros(row, col, CV_64FC1); //梯度幅值
59+
Mat thead = Mat::zeros(row, col, CV_64FC1); //梯度角度
60+
Mat whereGrad = Mat::zeros(row, col, CV_64FC1);//区域
61+
//x方向的Sobel算子
62+
Mat Sx = (cv::Mat_<double>(3, 3) << -1, 0, 1, -2, 0, 2, -1, 0, 1);
63+
//y方向的Sobel算子
64+
Mat Sy = (cv::Mat_<double >(3, 3) << 1, 2, 1, 0, 0, 0, -1, -2, -1);
65+
//计算梯度赋值和角度
66+
for(int i=1; i < row-1; i++){
67+
for(int j=1; j < col-1; j++){
68+
rect.x = j-1;
69+
rect.y = i-1;
70+
rect.width = 3;
71+
rect.height = 3;
72+
Mat rectImg = Mat::zeros(3, 3, CV_64FC1);
73+
filterImg(rect).copyTo(rectImg);
74+
//梯度和角度
75+
gradX.at<double>(i, j) += cv::sum(rectImg.mul(Sx)).val[0];
76+
gradY.at<double>(i, j) += cv::sum(rectImg.mul(Sy)).val[0];
77+
grad.at<double>(i, j) = sqrt(pow(gradX.at<double>(i, j), 2) + pow(gradY.at<double>(i, j), 2));
78+
thead.at<double>(i, j) = atan(gradY.at<double>(i, j)/gradX.at<double>(i, j));
79+
if(0 <= thead.at<double>(i, j) <= (PI/4.0)){
80+
whereGrad.at<double>(i, j) = 0;
81+
}else if(PI/4.0 < thead.at<double>(i, j) <= (PI/2.0)){
82+
whereGrad.at<double>(i, j) = 1;
83+
}else if(-PI/2.0 <= thead.at<double>(i, j) <= (-PI/4.0)){
84+
whereGrad.at<double>(i, j) = 2;
85+
}else if(-PI/4.0 < thead.at<double>(i, j) < 0){
86+
whereGrad.at<double>(i, j) = 3;
87+
}
88+
}
89+
}
90+
//printf("success\n");
91+
//梯度归一化
92+
double gradMax;
93+
cv::minMaxLoc(grad, &gradMax);
94+
if(gradMax != 0){
95+
grad = grad / gradMax;
96+
}
97+
//双阈值确定
98+
cv::Mat caculateValue = cv::Mat::zeros(row, col, CV_64FC1); //grad变成一维
99+
resize(grad, caculateValue, Size(1, grad.rows * grad.cols));
100+
cv::sort(caculateValue, caculateValue, CV_SORT_EVERY_COLUMN + CV_SORT_ASCENDING);//升序
101+
long long highIndex= row * col * hightThres;
102+
double highValue = caculateValue.at<double>(highIndex, 0); //最大阈值
103+
double lowValue = highValue * lowThres;
104+
//NMS
105+
for(int i = 1 ; i < row-1; i++ ){
106+
for( int j = 1; j<col-1; j++){
107+
// 八个方位
108+
double N = grad.at<double>(i-1, j);
109+
double NE = grad.at<double>(i-1, j+1);
110+
double E = grad.at<double>(i, j+1);
111+
double SE = grad.at<double>(i+1, j+1);
112+
double S = grad.at<double>(i+1, j);
113+
double SW = grad.at<double>(i-1, j-1);
114+
double W = grad.at<double>(i, j-1);
115+
double NW = grad.at<double>(i -1, j -1); // 区域判断，线性插值处理
116+
double tanThead; // tan角度
117+
double Gp1; // 两个方向的梯度强度
118+
double Gp2; // 求角度，绝对值
119+
tanThead = abs(tan(thead.at<double>(i,j)));
120+
switch ((int)whereGrad.at<double>(i,j)) {
121+
case 0: Gp1 = (1- tanThead) * E + tanThead * NE; Gp2 = (1- tanThead) * W + tanThead * SW; break;
122+
case 1: Gp1 = (1- tanThead) * N + tanThead * NE; Gp2 = (1- tanThead) * S + tanThead * SW; break;
123+
case 2: Gp1 = (1- tanThead) * N + tanThead * NW; Gp2 = (1- tanThead) * S + tanThead * SE; break;
124+
case 3: Gp1 = (1- tanThead) * W + tanThead *NW; Gp2 = (1- tanThead) * E + tanThead *SE; break;
125+
default: break;
126+
}
127+
// NMS -非极大值抑制和双阈值检测
128+
if(grad.at<double>(i, j) >= Gp1 && grad.at<double>(i, j) >= Gp2){
129+
//双阈值检测
130+
if(grad.at<double>(i, j) >= highValue){
131+
grad.at<double>(i, j) = highValue;
132+
dst.at<double>(i, j) = 255;
133+
} else if(grad.at<double>(i, j) < lowValue){
134+
grad.at<double>(i, j) = 0;
135+
} else{
136+
grad.at<double>(i, j) = lowValue;
137+
}
138+
} else{
139+
grad.at<double>(i, j) = 0;
140+
}
141+
}
142+
}
143+
//抑制孤立低阈值点 3*3. 找到高阈值就255
144+
for(int i = 1; i < row - 1; i++){
145+
for(int j = 1; j < col - 1; j++){
146+
if(grad.at<double>(i, j) == lowValue){
147+
//3*3 区域强度
148+
rect.x = i-1;
149+
rect.y = j-1;
150+
rect.width = 3;
151+
rect.height = 3;
152+
for(int x = 0; x < 3; x++){
153+
for(int y = 0; y < 3; y++){
154+
if(grad(rect).at<double>(x, y)==highValue){
155+
dst.at<double>(i, j) = 255;
156+
break;
157+
}
158+
}
159+
}
160+
}
161+
}
162+
}
163+
return dst;
164+
}
165+
166+
int main(){
167+
Mat src = imread("../lena.jpg");
168+
Mat gray;
169+
cvtColor(src, gray, CV_BGR2GRAY);
170+
Mat dst = CannyEdgeDetection(gray, 3, 0.8, 0.5);
171+
imshow("origin", src);
172+
imshow("gray", gray);
173+
imshow("result", dst);
174+
waitKey(0);
175+
imwrite("../result.jpg", gray);
176+
imwrite("../result2.jpg", dst);
177+
return 0;
178+
}

‎Feature Extraction/README.md

Lines changed: 3 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,3 +1,5 @@
11
# OpenCV和C++实现图像特征提取的一些算法
22

3-
- Sobel Edge Detection.cpp C++实现了OpenCV中的Sobel边缘检测算法。算法原理请看：https://blog.csdn.net/just_sort/article/details/85015054
3+
- Sobel Edge Detection.cpp C++实现了OpenCV中的Sobel边缘检测算法。算法原理请看：https://blog.csdn.net/just_sort/article/details/85015054
4+
5+
- Canny Edge Detection.cpp C++实现了OpenCV中的Canny边缘检测算法，相对Sobel算子可以提取更细，更可靠的边缘。算法原理请看：https://blog.csdn.net/just_sort/article/details/85053157

0 commit comments

Comments
 (0)