图像处理（四）：边缘检测（二）：sobel和canny

具体原理，参考博客点击打开链接，最常用的边缘检测算子有：sobel、canny、prewitt、Roberts等，sobel算子模板为：

sobel算子考虑了邻域信息，相当于对图像先做加权平均，然后再做微分运算。

Canny算子：

1、使用Gassian滤波器平滑处理；

2、使用sobel算子分别计算x，y方向梯度分量及梯度方向；

3、对梯度进行非极大值抑制；

4、使用双阈值对边缘进一步检测和增强；

实现代码如下：

#include   
#include   
#include   

using namespace cv;
using namespace std;

// define sobel  kernel  
const int sobel_kernel_y[9] = { -1, -2, -1, 0, 0, 0, 1, 2, 1 };
const int sobel_kernel_x[9] = { -1, 0, 1, -2, 0, 2, -1, 0, 1 };

// 3x3 位置  
const int locations[9][2] = { { -1, -1 }, { -1, 0 }, { -1, 1 },
{ 0, -1 }, { 0, 0 }, { 0, 1 },
{ 1, -1 }, { 1, 0 }, { 1, 1 } };

// 8领域  
const int neighbors[8][2] = { { -1, -1 }, { -1, 0 }, { -1, 1 }, { 0, -1 }, { 0, 1 }, { 1, -1 }, { 1, 0 }, { 1, 1 } };


int main()
{

	Mat src = imread("lena.jpg");
	Mat image = src.clone();
	GaussianBlur(src, image, Size(3, 3), 1.5);

	Mat gray;
	cvtColor(image, gray, CV_BGR2GRAY);


	Mat sobel_x = Mat::zeros(gray.rows, gray.cols, CV_8UC1);
	Mat sobel_y = Mat::zeros(gray.rows, gray.cols, CV_8UC1);
	Mat sobel_xy = Mat::zeros(gray.rows, gray.cols, CV_64FC1);

	int i, j, k;
	// compute sobel_x, sobel_y  
	for (i = 1; i < gray.rows - 1; i++)
		for (j = 1; j < gray.cols - 1; j++)
		{
			int temp_x = 0, temp_y = 0;
			for (k = 0; k < 9; k++)
			{
				temp_x += sobel_kernel_x[k] * gray.at(i + locations[k][0], j + locations[k][1]);
				temp_y += sobel_kernel_y[k] * gray.at(i + locations[k][0], j + locations[k][1]);
			}

			sobel_x.at(i, j) = temp_x;
			sobel_y.at(i, j) = temp_y;
			//sobel_xy.at(i, j) = sqrt(temp_x * temp_x + temp_y * temp_y);
			sobel_xy.at(i, j) = abs(temp_x) + abs(temp_y);
		}

	Mat directions = Mat::zeros(gray.rows, gray.cols, CV_64FC1);
	// compute direction  
	for (i = 1; i < gray.rows - 1; i++)
		for (j = 1; j < gray.cols - 1; j++)
		{
			// atan2的取值范围为[-pi,pi]  
			float t = atan2(sobel_y.at(i, j), sobel_x.at(i, j));
			if (t < 0)
			{
				t += CV_PI;
			}
			directions.at(i, j) = t;
		}

	float t = 0;
	// 非极大值抑制  
	for (i = 1; i < gray.rows - 1; i++)
		for (j = 1; j < gray.cols - 1; j++)
		{

			t = directions.at(i, j);
			// 0 - 22.5,   
			if (((t >= 0) && (t < CV_PI / 8.0)) || ((t >= 7.0 * CV_PI / 8.0) && (t < CV_PI)))
			{
				if ((sobel_xy.at(i, j) < sobel_xy.at(i, j + 1)) ||
					(sobel_xy.at(i, j) < sobel_xy.at(i, j - 1)))
				{
					sobel_xy.at(i, j) = 0;
				}
			}
			// 22.5 - 67.5  
			else if ((t >= CV_PI / 8.0) && (t < 3.0 * CV_PI / 8.0))
			{
				if ((sobel_xy.at(i, j) < sobel_xy.at(i - 1, j + 1)) ||
					(sobel_xy.at(i, j) < sobel_xy.at(i + 1, j - 1)))
				{
					sobel_xy.at(i, j) = 0;
				}
			}
			// 67.5 - 112.5  
			else if ((t >= 3.0 * CV_PI / 8.0) && (t < 5.0 * CV_PI / 8.0))
			{
				if ((sobel_xy.at(i, j) < sobel_xy.at(i - 1, j)) ||
					(sobel_xy.at(i, j) < sobel_xy.at(i + 1, j)))
				{
					sobel_xy.at(i, j) = 0;
				}
			}
			// 112.5 - 157.5  
			else if ((t >= 5.0 * CV_PI / 8.0) && (t < 7.0 * CV_PI / 8.0))
			{
				if ((sobel_xy.at(i, j) < sobel_xy.at(i - 1, j - 1)) ||
					(sobel_xy.at(i, j) < sobel_xy.at(i + 1, j + 1)))
				{
					sobel_xy.at(i, j) = 0;
				}
			}
		}

	// 双阈值滤波  
	float lower_t = 30;
	float upper_t = 100;

	Mat My_canny = Mat::zeros(sobel_xy.rows, sobel_xy.cols, CV_8UC1);
	Mat sobel_xy_mask = Mat::zeros(sobel_xy.rows, sobel_xy.cols, CV_32FC1);
	sobel_xy.copyTo(sobel_xy_mask);

	// 将两个阈值能够判断的进行判断
	for (i = 1; i < sobel_xy_mask.rows - 1; i++)
		for (j = 1; j < sobel_xy_mask.cols - 1; j++)
		{
			if (sobel_xy_mask.at(i, j) > upper_t)
			{
				My_canny.at(i, j) = 255;
				sobel_xy_mask.at  (i, j) = 0;   // 用完就置0，方便后面的检测
			}
			if (sobel_xy_mask.at(i, j) < lower_t)
			{
				sobel_xy.at(i, j) = 0;
			}
		}

	// 对两个阈值中间的部分进行增强
	for (i = 1; i < sobel_xy_mask.rows - 1; i++)
		for (j = 1; j < sobel_xy_mask.cols - 1; j++)
		{
			
			if (sobel_xy_mask.at(i, j ) >= lower_t)
			{
				// 遍历八领域 ，看八邻域是否有255
				for (k = 0; k < 8; k++)
				{
					if (My_canny.at(i + neighbors[k][0], j + neighbors[k][1]) == 255)
					{
						My_canny.at(i, j) = 255;
						sobel_xy_mask.at(i, j) = 0;
						break;
					}
				}

			}
			
		}


	Mat canny_sys;
	Canny(gray, canny_sys, 30, 100);

	namedWindow("gray", 0);
	resizeWindow("sobel_x", 300, 300);
	namedWindow("sobel_x", 0);
	resizeWindow("sobel_y", 300, 300);
	namedWindow("sobel_y", 0);
	resizeWindow("sobel_xy", 300, 300);
	namedWindow("sobel_xy", 0);
	resizeWindow("sobel_xy", 300, 300);
	namedWindow("My_canny", 0);
	resizeWindow("My_canny", 300, 300);
	namedWindow("canny_sys", 0);
	resizeWindow("canny_sys", 300, 300);


	imshow("gray", gray);
	imshow("sobel_x", sobel_x);
	imshow("sobel_y", sobel_y);
	imshow("sobel_xy", sobel_xy);
	imshow("My_canny", My_canny);
	imshow("canny_sys", canny_sys);


	waitKey();




	return 0;
}

效果如下：

跟opencv自带的canny效果好像存在一定偏差，具体原因等过段时间研究一下源代码~