SWT C++ OPENCV 像素点文字检测

#include "SWT.h" #include "ConnectDomain.h" int cmp(const void *a , const void *b) { float* x=(float*)a; float* y=(float*)b; if(*x<*y)return -1; else return 1; } void CSWT::loadImage(const string path) { srcImage=imread(path); } void CSWT::loadImage(char* path) { srcImage=imread(path); } void CSWT::loadImage(const Mat& im) { im.copyTo(srcImage); } void CSWT::grayImage() { imshow("original", srcImage); cvtColor(srcImage, grayImage8U, CV_BGR2GRAY); grayImage8U.convertTo(grayImage32F, CV_32FC1); height=srcImage.rows; width=srcImage.cols; } void CSWT::computeGradient()//计算梯度方向角度 { Canny(grayImage8U, edgeImage, 30, 90); Sobel(grayImage32F, gradient_dx, grayImage32F.depth(), 1, 0); Sobel(grayImage32F, gradient_dy, grayImage32F.depth(), 0, 1); angle.create(grayImage32F.size(), grayImage32F.type()); for(int h=0; h<height; ++h) { for(int w=0; w<width; ++w) { if(edgeImage.at<uchar>(h,w)) angle.at<float>(h,w)=atan2(gradient_dy.at<float>(h,w), gradient_dx.at<float>(h,w))/*/PI*180+180*/; //atan2()返回值（-PI,PI) else angle.at<float>(h,w)=0.0f; } } } void CSWT::computeSWT() { Mat tmp_SWTMap(grayImage32F.size(), grayImage32F.type(), Scalar(INF)); float strokeWidth; int initialX,initialY; int nextX, nextY; int step; float initialTheta; int* rayX=new int[maxStrokeWidth]; int* rayY=new int[maxStrokeWidth]; int indexRay; float oppositeTheta; int strokePointsX[100000];//每条射线的起点坐标(注意此处射线起点可能过多) int strokePointsY[100000]; int indexStrokePoints=0;//记录有多少笔画边缘点，即射线起点 //第一次扫描 for(int h=0; h<height; ++h) { for(int w=0; w<width; ++w) { if(edgeImage.at<uchar>(h,w)) { initialX=w; initialY=h; initialTheta=angle.at<float>(initialY,initialX); indexRay=0; step=1; rayX[indexRay]=initialX; rayY[indexRay]=initialY; indexRay++; while (step<maxStrokeWidth) { nextX = (int)(initialX + cos(initialTheta) * searchDirection * step); nextY = (int)(initialY + sin(initialTheta) * searchDirection * step); step++; if(nextY<0||nextX<0||nextX>=width||nextY>=height) { break; } rayX[indexRay]=nextX; rayY[indexRay]=nextY; indexRay++; if(edgeImage.at<uchar>(nextY,nextX)) { oppositeTheta=angle.at<float>(nextY,nextX); //if(fabs(initialTheta+oppositeTheta)<PI/4) if (abs(abs(initialTheta - oppositeTheta) - PI) < PI/3) { strokeWidth=sqrt((nextX-initialX)*(nextX-initialX)+(nextY-initialY)*(nextY-initialY)); for(int i=0; i<indexRay; ++i)//给射线上的所有点赋值 { tmp_SWTMap.at<float>(rayY[i],rayX[i])=min(tmp_SWTMap.at<float>(rayY[i],rayX[i]),strokeWidth); } strokePointsX[indexStrokePoints]=initialX; strokePointsY[indexStrokePoints]=initialY; indexStrokePoints++; } break; } }//while } }//for(w) } for(int h=0; h<height; ++h) { for(int w=0; w<width; ++w) { if (tmp_SWTMap.at<float>(h,w)==INF) { tmp_SWTMap.at<float>(h,w)=0; } } } //第二次扫描，每条射线上大于射线中值的像素值赋为中值 int* arryMedian=new int[maxStrokeWidth]; int indexMedian=0; for(int i=0; i<indexStrokePoints; ++i) { step=1; indexRay=0; indexMedian=0; initialX=strokePointsX[i]; initialY=strokePointsY[i]; initialTheta=angle.at<float>(initialY,initialX); rayX[indexRay]=initialX; rayY[indexRay]=initialY; indexRay++; arryMedian[indexMedian++]=tmp_SWTMap.at<float>(initialY, initialX); while (step<maxStrokeWidth)//找同一条射线上所有笔画宽度 { nextX = (int)(initialX + cos(initialTheta) * searchDirection * step); nextY = (int)(initialY + sin(initialTheta) * searchDirection * step); step++; rayX[indexRay]=nextX; rayY[indexRay]=nextY; indexRay++; arryMedian[indexMedian++]=tmp_SWTMap.at<float>(nextY,nextX); if(edgeImage.at<uchar>(nextY, nextX)) break; } qsort(arryMedian,indexMedian,sizeof(float) , cmp); float medianVal = arryMedian[indexMedian/2]; for(int j=0; j<indexRay; ++j) { tmp_SWTMap.at<float>(rayY[j], rayX[j])=min(tmp_SWTMap.at<float>(rayY[j], rayX[j]),medianVal); } } for(int h=0; h<height; ++h) { for(int w=0; w<width; ++w) { if (tmp_SWTMap.at<float>(h,w)>maxStrokeWidth) { tmp_SWTMap.at<float>(h,w)=0; } } } imshow("swtMap", tmp_SWTMap); //dilate(tmp_SWTMap, tmp_SWTMap, cv::Mat(), cv::Point(-1,-1), 1); Mat ker_x=(Mat_<int>(1,3)<<1,1,1); Mat ker_y=(Mat_<int>(2,1)<<1,1); cv::dilate(tmp_SWTMap, tmp_SWTMap, ker_x, Point(-1,-1),1); cv::dilate(tmp_SWTMap, tmp_SWTMap, ker_y, Point(-1,-1),1); tmp_SWTMap.copyTo(SWTMap); delete []arryMedian; delete []rayX; delete []rayY; } void CSWT::computeConnDomain() { //注意：相连的区域不一定就是同一连通域 labelMap = connDomain(SWTMap, connCount); cout << "connCount = " << connCount << endl; imshow("labelMap",labelMap); } void CSWT::filterConnDomain() { //第一步：计算各个连通域的属性： //varianceSW：连通域笔画的方差。meanSW：连通域笔画的均值。 //aspectRatio：连通域高宽比。diameter：连通域直径。 //width：连通域宽。height：连通域高 connAttri=new connAttribute[connCount+1]; //初始化属性列表 for(int i=1; i<=connCount; ++i) { connAttri[i].varianceSW = 0.0; connAttri[i].meanSW = 0.0; connAttri[i].aspectRatio = 0.0; connAttri[i].areaRatio = 0.0; connAttri[i].diameter = 0.0; connAttri[i].totalSW = 0.0; connAttri[i].width = 0; connAttri[i].height = 0; connAttri[i].num = 0; connAttri[i].L = width; connAttri[i].U = height; connAttri[i].R = 0; connAttri[i].D = 0; connAttri[i].flag = true; } for(int h=0; h<height; ++h) { for(int w=0; w<width; ++w) { int k = labelMap.at<float>(h,w); if(k) { connAttri[k].num++; connAttri[k].totalSW+=SWTMap.at<float>(h,w); if(connAttri[k].L>w)connAttri[k].L=w; if(connAttri[k].R<w)connAttri[k].R=w; if(connAttri[k].U>h)connAttri[k].U=h; if(connAttri[k].D<h)connAttri[k].D=h; } } } //计算每个连通域的属性 for(int i=1; i<=connCount; ++i) { connAttri[i].width =connAttri[i].R-connAttri[i].L+1; connAttri[i].height =connAttri[i].D-connAttri[i].U+1; connAttri[i].cx =(connAttri[i].L+connAttri[i].R)/2; connAttri[i].cy =(connAttri[i].U+connAttri[i].D)/2; connAttri[i].aspectRatio =(float)connAttri[i].width/connAttri[i].height; connAttri[i].areaRatio =(float)connAttri[i].num/(connAttri[i].width*connAttri[i].height); connAttri[i].meanSW =connAttri[i].totalSW/connAttri[i].num; connAttri[i].diameter =sqrt(connAttri[i].width*connAttri[i].width+connAttri[i].height*connAttri[i].height)/2; } //计算每个连通域的方差（近似计算） for(int h=0; h<height; ++h) { for(int w=0; w<width; ++w) { int k = labelMap.at<float>(h,w); if(k) { connAttri[k].varianceSW+=fabs(SWTMap.at<float>(h,w)-connAttri[k].meanSW); } } } Mat temp; labelMap.copyTo(temp); int num =0; for(int h=0; h<height; ++h) { for(int w=0; w<width; ++w) { int k = labelMap.at<float>(h,w); if(!k)continue; if(connAttri[k].width<10||connAttri[k].height<15||connAttri[k].width>(width/3)||connAttri[k].height>(height/2) ||connAttri[k].num<30||connAttri[k].num>1600 ||connAttri[k].aspectRatio<0.1||connAttri[k].aspectRatio>10.0 ||connAttri[k].areaRatio<0.35||connAttri[k].meanSW>20|| connAttri[k].varianceSW>6000 //||connAttri[k].diameter/connAttri[k].meanSW>10.0 ) { temp.at<float>(h,w)=0.0f; connAttri[k].flag = false; } } } for(int i=1; i<=connCount; ++i) { if(connAttri[i].flag) { num++