作为代码界的菜鸟,最近在尝试着用vs实现《数字图像处理与机器视觉》一书中有关图像处理的VC++代码。目前在研究空间域增强,希望自己能够越走越远^_^
本篇是图像锐化中高提升滤波的实现。
高提升滤波是针对Sobel、Laplacian等滤波的缺点提出的一种改进后的滤波,它的步骤主要是:
1)选择锐化算子对图像进行锐化
2)将原图像和锐化后的图像按照一定比例混合
3)对混合后的图像进行灰度调整,使混合后图像的灰度值落在[0,255]间
代码如下:
#include <iostream>
#include <opencv2\core\core.hpp>
#include <opencv2\highgui\highgui.hpp>
#include <opencv2\imgproc\imgproc.hpp>
using namespace std;
using namespace cv;
//常用滤波模板数组
//平均平滑1/9
float Template_Smooth_Avg[9] = { 1, 1, 1, 1, 1, 1, 1, 1, 1 };
//Gauss平滑1/16
float Template_Smooth_Gauss[9] = { 1, 2, 1, 2, 4, 2, 1, 2, 1 };
//Sobel垂直边缘检测
float Template_Smooth_HSobel[9] = { -1, 0, 1, -2, 0, 2, -1, 0, 1 };
//Sobel水平边缘检测
float Template_Smooth_VSobel[9] = { -1, -2, -1, 0, 0, 0, 1, 2, 1 };
//LOG边缘检测
float Template_Log[25] = { 0, 0, -1, 0, 0, 0, -1, -2, -1, 0, -1, -2, 16, -2, -1, 0, -1, -2, -1, 0, 0, 0, -1, 0, 0 };
//Laplacian边缘检测
float Template_Laplacian1[9] = { 0, -1, 0, -1, 4, -1, 0, -1, 0 };//对90度各向同性
float Template_Laplacian2[9] = { -1, -1, -1, -1, 8, -1, -1, -1, -1 };//对45度各向同性
/*************************************************************************************************************
高提升滤波
dProportion:高提升滤波中原图像的混合比例
nTempH:模板高度,nTempW:模板宽度
nTempMY:模板中心元素坐标,nTempMX:模板中心元素坐标
fpArray:指向模板数组的指针,可以选取不同模板实现不同滤波的高提升版本
fCoef:模板系数
**************************************************************************************************************/
void EnhanceFilter(Mat img, Mat &dst,double dProportion, int nTempH, int nTempW, int nTempMY, int nTempMX, float *pfArray, float fCoef)
{
int i, j, nHeight = img.rows, nWidth = img.cols;
vector<vector<int>> GrayMat1, GrayMat2, GrayMat3;//暂存按比例叠加图像,R,G,B三通道
vector<int> vecRow1(nWidth, 0), vecRow2(nWidth, 0), vecRow3(nWidth, 0);
for (i = 0; i < nHeight; i++)
{
GrayMat1.push_back(vecRow1);
GrayMat2.push_back(vecRow2);
GrayMat3.push_back(vecRow3);
}
//锐化图像,输出带符号响应,并与原图像按比例叠加
for (i = nTempMY; i < nHeight - (nTempH - nTempMY) + 1; i++)
{
for (j = nTempMX; j < nWidth - (nTempW - nTempMX) + 1; j++)
{
float fResult1 = 0;
float fResult2 = 0;
float fResult3 = 0;
for (int k = 0; k < nTempH; k++)
{
for (int l = 0; l < nTempW; l++)
{
//分别计算三通道加权和
fResult1 += img.at<Vec3b>(i, j)[0] * pfArray[k*nTempW + 1];
fResult2 += img.at<Vec3b>(i, j)[1] * pfArray[k*nTempW + 1];
fResult3 += img.at<Vec3b>(i, j)[2] * pfArray[k*nTempW + 1];
}
}
//三通道加权和分别乘以系数并限制响应范围,最后和原图像按比例混合
fResult1 *= fCoef;
if (fResult1>255)
fResult1 = 255;
if (fResult1 < -255)
fResult1 = -255;
GrayMat1[i][j] = dProportion*img.at<Vec3b>(i, j)[0] + fResult1 + 0.5;
fResult2 *= fCoef;
if (fResult2>255)
fResult2 = 255;
if (fResult2 < -255)
fResult2 = -255;
GrayMat2[i][j] = dProportion*img.at<Vec3b>(i, j)[1] + fResult2 + 0.5;
fResult3 *= fCoef;
if (fResult3>255)
fResult3 = 255;
if (fResult3 < -255)
fResult3 = -255;
GrayMat3[i][j] = dProportion*img.at<Vec3b>(i, j)[2] + fResult3 + 0.5;
}
}
int nMax1 = 0, nMax2 = 0, nMax3 = 0;//三通道最大灰度和值
int nMin1 = 65535, nMin2 = 65535, nMin3 = 65535;//三通道最小灰度和值
//分别统计三通道最大值最小值
for (i = nTempMY; i < nHeight - (nTempH - nTempMY) + 1; i++)
{
for (j = nTempMX; j < nWidth - (nTempW - nTempMX) + 1; j++)
{
if (GrayMat1[i][j]>nMax1)
nMax1 = GrayMat1[i][j];
if (GrayMat1[i][j] < nMin1)
nMin1 = GrayMat1[i][j];
if (GrayMat2[i][j]>nMax2)
nMax2 = GrayMat2[i][j];
if (GrayMat2[i][j] < nMin2)
nMin2 = GrayMat2[i][j];
if (GrayMat3[i][j]>nMax3)
nMax3 = GrayMat3[i][j];
if (GrayMat3[i][j] < nMin3)
nMin3 = GrayMat3[i][j];
}
}
//将按比例叠加后的三通道图像取值范围重新归一化到[0,255]
int nSpan1 = nMax1 - nMin1, nSpan2 = nMax2 - nMin2, nSpan3 = nMax3 - nMin3;
for (i = nTempMY; i < nHeight - (nTempH - nTempMY) + 1; i++)
{
for (j = nTempMX; j < nWidth - (nTempW - nTempMX) + 1; j++)
{
int br,bg,bb;
if (nSpan1>0)
br = (GrayMat1[i][j] - nMin1) * 255 / nSpan1;
else if (GrayMat1[i][j] <= 255)
br = GrayMat1[i][j];
else
br = 255;
dst.at<Vec3b>(i, j)[0] = br;
if (nSpan2>0)
bg = (GrayMat2[i][j] - nMin2) * 255 / nSpan2;
else if (GrayMat2[i][j] <= 255)
bg = GrayMat2[i][j];
else
bg = 255;
dst.at<Vec3b>(i, j)[1] = bg;
if (nSpan3>0)
bb = (GrayMat3[i][j] - nMin3) * 255 / nSpan3;
else if (GrayMat3[i][j] <= 255)
bb = GrayMat3[i][j];
else
bb = 255;
dst.at<Vec3b>(i, j)[2] = bb;
}
}
}
void main()
{
Mat img = imread("2.jpg");
cvNamedWindow("原图", CV_WINDOW_FREERATIO);
imshow("原图", img);
Mat dst=img.clone();
EnhanceFilter(img, dst, 1.8, 3, 3, 1, 1, Template_Laplacian2, 1);
cvNamedWindow("Laplacian锐化", CV_WINDOW_FREERATIO);
imshow("Laplacian锐化", dst);
waitKey(0);
}
原图:
结果图:
通过在main函数里调用Sobel函数,我尝试了高提升Sobel滤波,下面是核心代码和结果:
void FilterSobel(Mat img, Mat &dst)
{
Mat test1 = dst.clone();
Mat test2 = dst.clone();
EnhanceFilter(img, test1, 1.8, 3, 3, 1, 1, Template_Smooth_HSobel, 1);
EnhanceFilter(img, test2, 1.8, 3, 3, 1, 1, Template_Smooth_VSobel, 1);
dst = test1 + test2;
}
main函数里只要调用就可以了:
FilterSobel(img, dst);
结果图如下:
大家也可以尝试其他模板类型和大小。