下面我們對一幅人體骨骼掃描圖進行混合空間增強。之前在網上找了很多關於空間增強的代碼,但是都只有matlab的版本,一直沒有找到C++的版本。所以我結合岡薩雷斯的《數字圖像處理》上面的思路,粗略的寫了一個C++的版本。整個算法的框架如下:
#include<opencv2/opencv.hpp>
#include<iostream>
using namespace std;
using namespace cv;
void toBeOne(Mat &input, Mat &output, int index = 0)
{
float max = 0, min = 0;
output = Mat::zeros(input.rows, input.cols, CV_8UC3);
for (int i = 0; i<input.rows; i++)
{
float *ptr = input.ptr<float>(i);
for (int j = 0; j<input.cols * 3; j++)//圖像爲3通道
{
if (max<ptr[j])
max = ptr[j];
if (min>ptr[j])
min = ptr[j];
}
}
//取出圖像中的上下限
for (int i = 0; i<input.rows; i++)
{
float *ptr = input.ptr<float>(i);
uchar *optr = output.ptr<uchar>(i);
for (int j = 0; j<input.cols * 3; j++)
{
if (index){
if (ptr[j]>1){
optr[j] = 255;
continue;
}
else if (ptr[j]<0){
optr[j] = 0;
continue;
}
else
optr[j] = (uchar)(ptr[j] * 255);
}
else
optr[j] = (uchar)((ptr[j] - min) / (max - min) * 255);
}
}
}
//γ變換
cv::Mat gammaTran(const cv::Mat src, double gamma, double comp)
{
cv::Mat dst(src);
int M = 0;
int N = 0;
if (src.empty()){
std::cout << "Src pic is empty" << std::endl;
return src;
}
M = src.rows;
N = src.cols*src.channels();
for (int i = 0; i < M; i++){
const float *p1 = src.ptr<const float>(i);
float *p2 = dst.ptr<float>(i);
for (int j = 0; j < N; j++){
p2[j] = pow(p1[j], gamma) * comp;
}
}
return dst;
}
int main()
{
Mat iinput = imread("C:/Users/53148/Desktop/body1.png"), input, Tlaplas;
imshow("original", iinput);
iinput.convertTo(input, CV_32F, 1.0 / 255, 0);//把圖片轉化爲float類型,這樣子可以直接進行加減而不會溢出
//Laplacian變換
Mat kern = (Mat_<char>(3, 3) << 1, 1, 1, //濾波器
1, -8, 1,
1, 1, 1);
Mat laplas;
Mat output;
filter2D(input, laplas, input.depth(), kern);//使用濾波器kern對input進行相關操作,結果存儲在laplas中
toBeOne(laplas, Tlaplas);
imshow("Tlaplas", Tlaplas);
output = input - laplas;//如果中間的值是正的則是加號,負值則是減號
toBeOne(output, iinput, 1);
Mat R0 = iinput;
imshow("laplace", R0);
//Sobel梯度
Mat kern2 = (Mat_<char>(3, 3) << -1, -2, -1,
0, 0, 0,
1, 2, 1);
Mat kern3 = (Mat_<char>(3, 3) << -1, 0, 1,
-2, 0, 2,
-1, 0, 1);
Mat gx, gy;
filter2D(input, gx, input.depth(), kern2);
filter2D(input, gy, input.depth(), kern3);
Mat Soutput = abs(gx) + abs(gy);
toBeOne(Soutput, iinput, 1);
Mat R1 = iinput;
imshow("Sobel1", R1);
//盒濾波器平滑
Mat smooth;
blur(iinput, smooth, Size(5, 5));
Mat R2 = smooth;
imshow("Smooth", R2);
//相乘
Mat R3, r3;
R0.convertTo(R0, CV_32F, 1.0 / 255, 0);
R2.convertTo(R2, CV_32F, 1.0 / 255, 0);
multiply(R0, R2, r3);
toBeOne(r3, R3, 1);
imshow("Result0",R3);
//銳化增強
cv::Mat r4, R4;
r4 = input + r3;
toBeOne(r4, R4, 1);
imshow("Result1", R4);
//γ變換
cv::Mat r5, R5;
r5 = gammaTran(r4, 0.5, 1);
toBeOne(r5, R5, 1);
imshow("Result", R5);
waitKey();
return 0;
}
通過增強的效果圖如下:
以上僅僅是我學習過程中的筆記,如有什麼疏漏,敬請指正!