基于梯度方向、极化变换和聚类算法的图像主特征直线检测
基于机器学习和图像处理的直线检测
代码主要思路：
1）借助类LSD直线检测，提取图像各个方向梯度；
2）对像素中的各个梯度方向做极化变换；
3）对计划变换后的结果进行聚类运算；
4）对聚类运算结果，选取topK，进行极化反变换；
5）对结果进行腐蚀+膨胀+滤波，进行结果整合；
6）可根据图像的长度、斜率等进行配置筛选和去重。
代码结果以及下载地址：下载地址
主函数代码
#include <iostream>
#include <opencv2/core/core.hpp>
#include <opencv2/highgui/highgui.hpp>
#include <opencv2/imgproc/imgproc.hpp>
#include <string>
#include <vector>
#include "lsd.h"
#include <fstream>
#include <math.h>
#include <algorithm>
#include <Cstring>
#include <io.h>
#include<numeric>
#define PI 3.1415926
using namespace std;
using namespace cv;
typedef struct {
image_double imgDouble;
Mat imgSrc, imgPro;
}imageInfo;
void listFiles(const char * dir);
imageInfo imageProcess(char * DIR, double scale);
int lineLengthThreshold(ntuple_list LS, double scale);
vector<float> getPolarLine(vector<float> p);
void connectLine(Mat imgPlot, vector<float> decarTopX0, vector<float> decarTopY0,
vector<float> decarTopX1, vector<float> decarTopY1);
void HighPassFilterSobel(const Mat& image, Mat& result);
void main()
{
//生成图像索引
char dir[20] = "images\\";
listFiles(dir);
ifstream imageIndex;
imageIndex.open("imageIndex.txt");
const int LINE = 200;
char str[LINE];
while (imageIndex.getline(str, LINE)) {
char DIR[260] = "images\\";
strcat_s(DIR, str);
double scale = 0.8;
imageInfo imgInfo = imageProcess(DIR, scale);
image_double image = imgInfo.imgDouble;
Mat img = imgInfo.imgPro;
Mat imgPlot = imgInfo.imgSrc;
imshow("imgPlot", imgPlot);
ntuple_list ls0 = lsd(image);
cout << "ls0->size:" << ls0->size << endl;
int binNum = 60;
vector<double> disNum(binNum, 0);
double binLength = img.rows / binNum;
//设置直线筛选长度
int lineThreshold = lineLengthThreshold(ls0, scale);
cout << "lineThreshold:" << lineThreshold << endl;
//将笛卡尔坐标系点转到极坐标系
vector<float> polarR, polarTheta, decarPointLocX0,
decarPointLocY0, decarPointLocX1, decarPointLocY1;
vector<float> indexStep;//记录累加的步长
for (int i = 0; i < ls0->size; i++) {
ls0->values[5 * i + 0] = (int)(ls0->values[5 * i + 0] / scale);
ls0->values[5 * i + 1] = (int)(ls0->values[5 * i + 1] / scale);
ls0->values[5 * i + 2] = (int)(ls0->values[5 * i + 2] / scale);
ls0->values[5 * i + 3] = (int)(ls0->values[5 * i + 3] / scale);
int x0 = ls0->values[5 * i + 0];
int y0 = ls0->values[5 * i + 1];
int x1 = ls0->values[5 * i + 2];
int y1 = ls0->values[5 * i + 3];
//线长筛选并记录
float lineLength = sqrt((x1 - x0)*(x1 - x0) + (y1 - y0)*(y1 - y0));
if (lineLength < lineThreshold) {
continue;
}
//边缘筛选
int xTemp = img.cols - 10;
int yTemp = img.rows - 10;
if ((x0 < 10 && x1 < 10) || (x0 > xTemp && x1 > xTemp) || (y0 < 10 && y1 < 10) || (y0 > yTemp && y1 > yTemp)) {
continue;
}
Point pp0 = Point(x0, y0);
Point pp1 = Point(x1, y1);
line(img, pp0, pp1, Scalar(225, 0, 0), 2, 4);
circle(img, pp0, 10, Scalar(0, 0, 0));
circle(img, pp1, 10, Scalar(255, 255, 255));
float tempIndexStep = lineLength / lineThreshold;
//cout << "indexStep:" << tempIndexStep << endl;
indexStep.push_back(tempIndexStep);
decarPointLocX0.push_back(x0);
decarPointLocY0.push_back(y0);
decarPointLocX1.push_back(x1);
decarPointLocY1.push_back(y1);
//统一坐标系并记录
vector<float> decarPoint;
int x0Point = x0 - img.cols / 2;
int y0Point = img.rows / 2 - y0;
int x1Point = x1 - img.cols / 2;
int y1Point = img.rows / 2 - y1;
decarPoint.push_back(x0Point);
decarPoint.push_back(y0Point);
decarPoint.push_back(x1Point);
decarPoint.push_back(y1Point);
//获取极坐标并记录
vector<float> polarPoint = getPolarLine(decarPoint);
polarR.push_back(polarPoint[0]);
polarTheta.push_back(polarPoint[1]);
}
float rThreshold = 15;
float thetaThreshold = 0.1;
//寻找各个值的聚类统计数量
int totalLine = polarR.size();
cout << "totalLine:" << totalLine << endl;
vector<float> index(totalLine, 0);
for (int i = 0; i < totalLine; i++) {
for (int j = 0; j < totalLine; j++) {
if (i == j) {
continue;
}
float subpolarR = abs(polarR[i] - polarR[j]);
float subpolarTheta = abs(polarTheta[i] - polarTheta[j]);
if (subpolarR < rThreshold && subpolarTheta < thetaThreshold) {
if (indexStep[j] != 0)
index[i] = index[i] + indexStep[j];
else
index[i]++;
}
}
}
//寻找聚类区域最多的点的索引TOP1
vector<float>::iterator maxIndex = max_element(index.begin(), index.end());
cout << "------------------------------" << endl;
cout << "max:" << *maxIndex << endl;
int maxInd = distance(index.begin(), maxIndex);
cout << "maxInd:" << maxInd << endl;
//存储聚类最多点的坐标并画线TOP1
vector<float> decarTop1X0, decarTop1Y0, decarTop1X1, decarTop1Y1;
vector<int> setLabel(totalLine, 0);//0：该点未使用；1：该点已使用
for (int i = 0; i < totalLine; i++) {
float subpolarR = abs(polarR[i] - polarR[maxInd]);
float subpolarTheta = abs(polarTheta[i] - polarTheta[maxInd]);
if (subpolarR < rThreshold && subpolarTheta < thetaThreshold) {
decarTop1X0.push_back(decarPointLocX0[i]);
decarTop1Y0.push_back(decarPointLocY0[i]);
decarTop1X1.push_back(decarPointLocX1[i]);
decarTop1Y1.push_back(decarPointLocY1[i]);
setLabel[i] = 1;
Point pp0 = Point(decarPointLocX0[i], decarPointLocY0[i]);
Point pp1 = Point(decarPointLocX1[i], decarPointLocY1[i]);
//line(imgPlot, pp0, pp1, Scalar(255, 0, 0), 3, 4);
}
}
connectLine(imgPlot, decarTop1X0, decarTop1Y0, decarTop1X1, decarTop1Y1);
//寻找聚类区域第二多的点的索引TOP2
int usedPointTop2 = 0;
vector<float> secondIndex(totalLine, 0);//剔除第一大的元素后的存储位置
for (int i = 0; i < totalLine; i++) {
if (setLabel[i] == 0) {
secondIndex[i] = index[i];
usedPointTop2++;
}
else
secondIndex[i] = 0;
//cout << secondIndex[i] << "-";
}
if (usedPointTop2 < 4) {
continue;
}
//cout << endl;
vector<float>::iterator secondMaxIndex = max_element(secondIndex.begin(), secondIndex.end());
cout << "------------------------------" << endl;
cout << "secondMax:" << *secondMaxIndex << endl;
int secondMaxInd = distance(secondIndex.begin(), secondMaxIndex);
cout << "secondMaxInd:" << secondMaxInd << endl;
//存储聚类第二多点的坐标并画线TOP2
vector<float> decarTop2X0, decarTop2Y0, decarTop2X1, decarTop2Y1;
for (int i = 0; i < totalLine; i++) {
float subpolarR = abs(polarR[i] - polarR[secondMaxInd]);
float subpolarTheta = abs(polarTheta[i] - polarTheta[secondMaxInd]);
if (subpolarR < rThreshold && subpolarTheta < thetaThreshold && setLabel[i] == 0) {
decarTop2X0.push_back(decarPointLocX0[i]);
decarTop2Y0.push_back(decarPointLocY0[i]);
decarTop2X1.push_back(decarPointLocX1[i]);
decarTop2Y1.push_back(decarPointLocY1[i]);
Point pp0 = Point(decarPointLocX0[i], decarPointLocY0[i]);
Point pp1 = Point(decarPointLocX1[i], decarPointLocY1[i]);
setLabel[i] = 1;
//line(imgPlot, pp0, pp1, Scalar(0, 255, 0), 3, 4);
}
}
connectLine(imgPlot, decarTop2X0, decarTop2Y0, decarTop2X1, decarTop2Y1);
//寻找聚类区域第三多的点的索引TOP3
int usedPointTop3 = 0;
vector<float> thirdIndex(totalLine, 0);//剔除第一大和第二大的元素后的存储位置
for (int i = 0; i < totalLine; i++) {
if (setLabel[i] == 0) {
thirdIndex[i] = secondIndex[i];
usedPointTop3++;
}
else
thirdIndex[i] = 0;
//cout << thirdIndex[i] << "-";
}
//cout << endl;
if (usedPointTop3 < 4) {
continue;
}
vector<float>::iterator thirdMaxIndex = max_element(thirdIndex.begin(), thirdIndex.end());
cout << "------------------------------" << endl;
cout << "thirdMax:" << *thirdMaxIndex << endl;
int thirdMaxInd = distance(thirdIndex.begin(), thirdMaxIndex);
cout << "thirdMaxInd:" << thirdMaxInd << endl;
//存储聚类第三多点的坐标并画线TOP3
vector<float> decarTop3X0, decarTop3Y0, decarTop3X1, decarTop3Y1;
for (int i = 0; i < totalLine; i++) {
float subpolarR = abs(polarR[i] - polarR[thirdMaxInd]);
float subpolarTheta = abs(polarTheta[i] - polarTheta[thirdMaxInd]);
if (subpolarR < rThreshold && subpolarTheta < thetaThreshold && setLabel[i] == 0) {
decarTop3X0.push_back(decarPointLocX0[i]);
decarTop3Y0.push_back(decarPointLocY0[i]);
decarTop3X1.push_back(decarPointLocX1[i]);
decarTop3Y1.push_back(decarPointLocY1[i]);
Point pp0 = Point(decarPointLocX0[i], decarPointLocY0[i]);
Point pp1 = Point(decarPointLocX1[i], decarPointLocY1[i]);
setLabel[i] = 1;
//line(imgPlot, pp0, pp1, Scalar(0, 0, 255), 3, 4);
}
}
connectLine(imgPlot, decarTop3X0, decarTop3Y0, decarTop3X1, decarTop3Y1);
//寻找聚类区域第四多的点的索引TOP4
int usedPointTop4 = 0;
vector<float> fourthIndex(totalLine, 0);//剔除第一大的元素后的存储位置
for (int i = 0; i < totalLine; i++) {
if (setLabel[i] == 0) {
fourthIndex[i] = thirdIndex[i];
usedPointTop4++;
}
else
fourthIndex[i] = 0;
//cout << fourthIndex[i] << "-";
}
//cout << endl;
vector<float>::iterator fourthMaxIndex = max_element(fourthIndex.begin(), fourthIndex.end());
cout << "------------------------------" << endl;
cout << "fourthMax:" << *fourthMaxIndex << endl;
int fourthMaxInd = distance(fourthIndex.begin(), fourthMaxIndex);
cout << "fourthMaxInd:" << fourthMaxInd << endl;
if (usedPointTop4 < 4 || *fourthMaxIndex == 0) {
continue;
}
//存储聚类第四多点的坐标并画线TOP4
vector<float> decarTop4X0, decarTop4Y0, decarTop4X1, decarTop4Y1;
for (int i = 0; i < totalLine; i++) {
float subpolarR = abs(polarR[i] - polarR[fourthMaxInd]);
float subpolarTheta = abs(polarTheta[i] - polarTheta[fourthMaxInd]);
if (subpolarR < rThreshold && subpolarTheta < thetaThreshold && setLabel[i] == 0) {
decarTop4X0.push_back(decarPointLocX0[i]);
decarTop4Y0.push_back(decarPointLocY0[i]);
decarTop4X1.push_back(decarPointLocX1[i]);
decarTop4Y1.push_back(decarPointLocY1[i]);
Point pp0 = Point(decarPointLocX0[i], decarPointLocY0[i]);
Point pp1 = Point(decarPointLocX1[i], decarPointLocY1[i]);
setLabel[i] = 1;
//line(img, pp0, pp1, Scalar(0, 255, 255), 3, 4);
}
}
connectLine(img, decarTop4X0, decarTop4Y0, decarTop4X1, decarTop4Y1);
//寻找聚类区域第五多的点的索引TOP5
int usedPointTop5 = 0;
vector<float> fifthIndex(totalLine, 0);//剔除第一大的元素后的存储位置
for (int i = 0; i < totalLine; i++) {
if (setLabel[i] == 0) {
fifthIndex[i] = fourthIndex[i];
usedPointTop5++;
}
else
fifthIndex[i] = 0;
//cout << fifthIndex[i] << "-";
}
//cout << endl;
vector<float>::iterator fifthMaxIndex = max_element(fifthIndex.begin(), fifthIndex.end());
cout << "------------------------------" << endl;
cout << "fifthMax:" << *fifthMaxIndex << endl;
int fifthMaxInd = distance(fifthIndex.begin(), fifthMaxIndex);
cout << "fifthMaxInd:" << fifthMaxInd << endl;
if (usedPointTop5 < 4 || *fifthMaxIndex == 0) {
continue;
}
//存储聚类第五多点的坐标并画线TOP5
vector<float> decarTop5X0, decarTop5Y0, decarTop5X1, decarTop5Y1;
for (int i = 0; i < totalLine; i++) {
float subpolarR = abs(polarR[i] - polarR[fifthMaxInd]);
float subpolarTheta = abs(polarTheta[i] - polarTheta[fifthMaxInd]);
if (subpolarR < rThreshold && subpolarTheta < thetaThreshold && setLabel[i] == 0) {
decarTop5X0.push_back(decarPointLocX0[i]);
decarTop5Y0.push_back(decarPointLocY0[i]);
decarTop5X1.push_back(decarPointLocX1[i]);
decarTop5Y1.push_back(decarPointLocY1[i]);
Point pp0 = Point(decarPointLocX0[i], decarPointLocY0[i]);
Point pp1 = Point(decarPointLocX1[i], decarPointLocY1[i]);
setLabel[i] = 1;
//line(img, pp0, pp1, Scalar(255, 0, 255), 3, 4);
}
}
connectLine(img, decarTop5X0, decarTop5Y0, decarTop5X1, decarTop5Y1);
char resultPath[200] = "resultPath//";
strcat_s(resultPath, str);
imwrite(resultPath, imgPlot);
namedWindow("IMG", CV_WINDOW_AUTOSIZE);
//resizeWindow("IMG", 600, 800);
imshow("IMG", imgPlot);
waitKey(1);
}
system("Pause");
}
转载：https://blog.csdn.net/cuihao1995/article/details/116614897
查看评论