OpenCV filter2D函数简介
OpenCV filter2D将图像与内核进行卷积,将任意线性滤波器应用于图像。支持就地操作。当孔径部分位于图像之外时,该函数根据指定的边界模式插值异常像素值。
该函数实际上计算相关性,而不是卷积:
filter2D函数的原型如下:
void cv::filter2D(InputArray src,
OutputArray dst,
int ddepth,
InputArray kernel,
Point anchor = Point(-1,-1),
double delta = 0,
int borderType = BORDER_DEFAULT
)
参数:
src 输入图像
dst 输出图像,与 src 大小相同、通道数相同
ddepth 目标图像的所需深度
kernel 卷积核(或者更确切地说是相关核),单通道浮点矩阵;如果要将不同的内核应用于 不同的通道,请使用 split 将图像分割为单独的颜色平面并单独处理它们。
anchor 内核的锚点,指示内核中过滤点的相对位置;锚应该位于内核内;默认值(-1,-1) 表示锚点位于内核中心。
delta 在将过滤像素存储到 dst 之前添加到过滤像素的可选值。
borderType 像素外推方法。可以选以下几种:BORDER_CONSTANT,BORDER_REPLICATE,BORDER_REFLECT,BORDER_REFLECT_101,BORDER_TRANSPARENT,BORDER_REFLECT101,BORDER_DEFAULT,BORDER_ISOLATED。
OpenCV filter2D函数应用
使用OpenCV filter2D函数,通过改变卷积核(kernel)可达成不同的滤波效果。下面就OpenCV filter2D函数的几种常用场景做说明,并以实例做演示。
图像锐化
图像锐化使用的卷积核如下:
Mat kernel = (Mat_<char>(3, 3) << 0, -1, 0, -1, 5, -1, 0, -1, 0);
下面以实例演示图像锐化操作及锐化效果,示例代码如下:
#include <iostream>
#include <opencv2/opencv.hpp>using namespace cv;
using namespace std;int main(int argc, char** argv)
{Mat src = imread("1.jpg");if (src.empty()){cout << "Cann't open Image" << endl;return -1;}imshow("Input Image", src);Mat kernel = (Mat_<char>(3, 3) << 0, -1, 0, -1, 5, -1, 0, -1, 0);Mat dst;filter2D(src, dst, src.depth(), kernel);imshow("Output Image", dst);waitKey(0);return 0;
}
试运行,结果如下:
可以看到经过Filter2D滤波后的图像变得更清晰。
均值滤波
OpenCV filter2D函数实现均值滤波的卷积核如下:
Mat kernel = (Mat_<float>(3, 3) << 1, 1, 1, 1, 1, 1, 1, 1, 1) / 9;
下面以实例演示filter2D实现图像均值滤波操作及滤波效果,示例代码如下:
#include <iostream>
#include <opencv2/opencv.hpp>using namespace cv;
using namespace std;int main(int argc, char** argv)
{//sharp test/*Mat src = imread("1.jpg");if (src.empty()){cout << "Cann't open Image" << endl;return -1;}imshow("Input Image", src);Mat kernel = (Mat_<char>(3, 3) << 0, -1, 0, -1, 5, -1, 0, -1, 0);Mat dst;filter2D(src, dst, src.depth(), kernel);imshow("Output Image", dst);*///Mean filter testMat src = imread("3.png");if (src.empty()){cout << "Cann't open Image" << endl;return -1;}imshow("Input Image", src);Mat kernel = (Mat_<float>(3, 3) << 1, 1, 1, 1, 1, 1, 1, 1, 1) / 9;Mat dst;filter2D(src, dst, src.depth(), kernel);for (size_t i = 0; i < 15; i++){filter2D(dst, dst, src.depth(), kernel);}imshow("Output Image", dst);waitKey(0);return 0;
}
试运行,结果如下:
可以看出,均值滤波可以去除图像椒盐噪声,达到磨皮效果。
高斯滤波
OpenCV filter2D函数实现高斯滤波的卷积核可由高斯核转换得到,方法如下:
Mat kernelGaussian = getGaussianKernel(9, 1.5);
Mat kernel = kernelGaussian * kernelGaussian.t();
下面以实例演示filter2D实现图像高斯滤波操作及滤波效果,示例代码如下:
#include <iostream>
#include <opencv2/opencv.hpp>using namespace cv;
using namespace std;int main(int argc, char** argv)
{//filter2d sharp test/*Mat src = imread("1.jpg");if (src.empty()){cout << "Cann't open Image" << endl;return -1;}imshow("Input Image", src);Mat kernel = (Mat_<char>(3, 3) << 0, -1, 0, -1, 5, -1, 0, -1, 0);Mat dst;filter2D(src, dst, src.depth(), kernel);imshow("Output Image", dst);*///filter2d Mean filter test/*Mat src = imread("3.png");if (src.empty()){cout << "Cann't open Image" << endl;return -1;}imshow("Input Image", src);Mat kernel = (Mat_<float>(3, 3) << 1, 1, 1, 1, 1, 1, 1, 1, 1) / 9;Mat dst;filter2D(src, dst, src.depth(), kernel);for (size_t i = 0; i < 15; i++){filter2D(dst, dst, src.depth(), kernel);}imshow("Output Image", dst);*///filter2d Gaussian filter testMat src = imread("3.png");if (src.empty()){cout << "Cann't open Image" << endl;return -1;}imshow("Input Image", src);Mat kernelGaussian = getGaussianKernel(9, 1.5);Mat kernel = kernelGaussian * kernelGaussian.t();Mat dst;filter2D(src, dst, src.depth(), kernel);for (size_t i = 0; i < 6; i++){filter2D(dst, dst, src.depth(), kernel);}imshow("Output Image", dst);waitKey(0);return 0;
}
试运行,结果如下:
可以看出,同样filter2D均高斯滤波同样可以去除图像椒盐噪声,达成磨皮效果,且所需次数更少。
边缘检测
filter2D还可以使用sobel内核实现边缘检测,soble内核如下:
Mat sobelX = (Mat_<float>(3, 3) << -1, 0, 1,-2, 0, 2,-1, 0, 1);
Mat sobelY = (Mat_<float>(3, 3) << -1, -2, -1,0, 0, 0,1, 2, 1);
下面以实例演示filter2D 用sobel核实现图像边缘检测操作及滤波效果,示例代码如下:
#include <iostream>
#include <opencv2/opencv.hpp>using namespace cv;
using namespace std;int main(int argc, char** argv)
{//filter2d sharp test/*Mat src = imread("1.jpg");if (src.empty()){cout << "Cann't open Image" << endl;return -1;}imshow("Input Image", src);Mat kernel = (Mat_<char>(3, 3) << 0, -1, 0, -1, 5, -1, 0, -1, 0);Mat dst;filter2D(src, dst, src.depth(), kernel);imshow("Output Image", dst);*///filter2d Mean filter test/*Mat src = imread("3.png");if (src.empty()){cout << "Cann't open Image" << endl;return -1;}imshow("Input Image", src);Mat kernel = (Mat_<float>(3, 3) << 1, 1, 1, 1, 1, 1, 1, 1, 1) / 9;Mat dst;filter2D(src, dst, src.depth(), kernel);for (size_t i = 0; i < 15; i++){filter2D(dst, dst, src.depth(), kernel);}imshow("Output Image", dst);*///filter2d Gaussian filter test/*Mat src = imread("3.png");if (src.empty()){cout << "Cann't open Image" << endl;return -1;}imshow("Input Image", src);Mat kernelGaussian = getGaussianKernel(9, 1.5);Mat kernel = kernelGaussian * kernelGaussian.t();Mat dst;filter2D(src, dst, src.depth(), kernel);for (size_t i = 0; i < 6; i++){filter2D(dst, dst, src.depth(), kernel);}imshow("Output Image", dst);*///filter2d detect edges testMat src = imread("4.png");if (src.empty()){cout << "Cann't open Image" << endl;return -1;}threshold(src, src, 127, 255, THRESH_BINARY);imshow("Input Image", src);Mat sobelX = (Mat_<float>(3, 3) << -1, 0, 1,-2, 0, 2,-1, 0, 1);Mat sobelY = (Mat_<float>(3, 3) << -1, -2, -1,0, 0, 0,1, 2, 1);Mat edges,edgesX, edgesY;filter2D(src, edgesX, CV_16S, sobelX);filter2D(src, edgesY, CV_16S, sobelX);convertScaleAbs(edgesX, edgesX);convertScaleAbs(edgesY, edgesY);addWeighted(edgesX, 0.5, edgesY, 0.5, 0, edges);imshow("Edges", edges);waitKey(0);return 0;
}
试运行,结果如下:
可以看出确实检测到了边缘,效果并不是很好。
filter2D还可以使用Prewitt核,实现边缘检测。Prewitt核如下:
Mat prewitt_x = (Mat_<int>(3, 3) << -1, 0, 1, -1, 0, 1, -1, 0, 1);
Mat prewitt_y = (Mat_<int>(3, 3) << -1, -1, -1,0, 0, 0, 1, 1, 1);
下面以实例演示filter2D 用Prewitt核实现图像边缘检测操作及滤波效果,示例代码如下:
#include <iostream>
#include <opencv2/opencv.hpp>using namespace cv;
using namespace std;int main(int argc, char** argv)
{//filter2d sharp test/*Mat src = imread("1.jpg");if (src.empty()){cout << "Cann't open Image" << endl;return -1;}imshow("Input Image", src);Mat kernel = (Mat_<char>(3, 3) << 0, -1, 0, -1, 5, -1, 0, -1, 0);Mat dst;filter2D(src, dst, src.depth(), kernel);imshow("Output Image", dst);*///filter2d Mean filter test/*Mat src = imread("3.png");if (src.empty()){cout << "Cann't open Image" << endl;return -1;}imshow("Input Image", src);Mat kernel = (Mat_<float>(3, 3) << 1, 1, 1, 1, 1, 1, 1, 1, 1) / 9;Mat dst;filter2D(src, dst, src.depth(), kernel);for (size_t i = 0; i < 15; i++){filter2D(dst, dst, src.depth(), kernel);}imshow("Output Image", dst);*///filter2d Gaussian filter test/*Mat src = imread("3.png");if (src.empty()){cout << "Cann't open Image" << endl;return -1;}imshow("Input Image", src);Mat kernelGaussian = getGaussianKernel(9, 1.5);Mat kernel = kernelGaussian * kernelGaussian.t();Mat dst;filter2D(src, dst, src.depth(), kernel);for (size_t i = 0; i < 6; i++){filter2D(dst, dst, src.depth(), kernel);}imshow("Output Image", dst);*///filter2d detect edges test/*//sobel kernelMat src = imread("4.png");if (src.empty()){cout << "Cann't open Image" << endl;return -1;}threshold(src, src, 127, 255, THRESH_BINARY);imshow("Input Image", src);Mat sobelX = (Mat_<float>(3, 3) << -1, 0, 1,-2, 0, 2,-1, 0, 1);Mat sobelY = (Mat_<float>(3, 3) << -1, -2, -1,0, 0, 0,1, 2, 1);Mat edges,edgesX, edgesY;filter2D(src, edgesX, CV_16S, sobelX);filter2D(src, edgesY, CV_16S, sobelX);convertScaleAbs(edgesX, edgesX);convertScaleAbs(edgesY, edgesY);addWeighted(edgesX, 0.5, edgesY, 0.5, 0, edges);imshow("Edges", edges);*///Prewitt kernelMat src = imread("4.png");if (src.empty()){cout << "Cann't open Image" << endl;return -1;}threshold(src, src, 127, 255, THRESH_BINARY);imshow("Input Image", src);Mat prewitt_x = (Mat_<int>(3, 3) << -1, 0, 1, -1, 0, 1, -1, 0, 1);Mat prewitt_y = (Mat_<int>(3, 3) << -1, -1, -1,0, 0, 0, 1, 1, 1);Mat edges, edgesX, edgesY;filter2D(src, edgesX, src.depth(), prewitt_x);filter2D(src, edgesY, src.depth(), prewitt_y);addWeighted(edgesX, 0.5, edgesY, 0.5, 0, edges);imshow("Edges", edges);waitKey(0);return 0;
}
试运行,结果如下:
从结果可以看出,filter2D使用Prewitt核检测边缘的结果,与使用sobel核边缘检测的结果是有差异的。
OpenCV filter2D函数就介绍到这里。博文示例是基于OpenCV4.8(opencv目录位于d盘根目录下)及VS2022。示例源码已上传到CSDN,其链接为:https://mp.csdn.net/mp_blog/creation/editor/136590730