Sobel算子取代：基于特定点方向的canny边缘检测
Posted wishchin
tags:
篇首语：本文由小常识网(cha138.com)小编为大家整理，主要介绍了Sobel算子取代：基于特定点方向的canny边缘检测相关的知识，希望对你有一定的参考价值。
前言：
Canny边缘检测使用了Sobel算子，计算dx和dy两个方向，对于特定方向的边缘检测，可以作少量修改。
代码：
计算特定方向上的边缘
void CannyOrient(
cv::Mat &_src, cv::Mat &_dst,
cv::Point2f &seed,
double low_thresh, double high_thresh,
int aperture_size, bool L2gradient )
low_thresh = 0;
high_thresh = 1;
//使用特定种子方向上的点寻找方向梯度
FindBestGradient( _src, _dst,seed,aperture_size, L2gradient );
//二值化，生成边缘图像
//计算直方图，寻找极大极小边缘点，根据波峰和波谷
int hh[256];
memset(hh,0,256*sizeof(int ) );
int minV = 255;
int maxV = 0;
cvWish::CalHist(_dst,1,minV,maxV,hh);
cv::Mat canvas( 600 ,256*4, CV_8U, 1);
for(int i=0;i< 256; ++i)
int Pos = ( (int)(hh[i]) ) %( canvas.rows);
canvas.at<uchar>( canvas.rows - Pos-1,i*4 ) =255;
cv::imshow("canvas",canvas);
cv::waitKey(1);
return;
寻找特定边缘：
//在种子点方向上寻找合适的梯度，用于寻找边缘
void FindBestGradient(
cv::Mat &_src, cv::Mat &_dst,
cv::Point2f &seed,
int aperture_size, bool L2gradient )
//角度矩阵
cv::Mat df = cv::Mat::zeros( _src.rows,_src.cols, CV_32FC1 );
//梯度矩阵
cv::Mat dg = cv::Mat::zeros( _src.rows,_src.cols, CV_32FC1 );
//原始图像
cv::Mat ds = _src.clone();
//目标图像 uchar型
cv::Mat dd = _src.clone();
//1.根据角度计算梯度//得到梯度矩阵
//使用N*1的算子
int n = aperture_size;//必须为奇数
//对每个柱进行初始化
//搜索柱:在射线方向上搜索l_Search 个像素；宽度为
int l_Search = n;
int w_Search = 1;
std::vector<std::vector<std::pair<cv::Point ,float> > > beam;
beam.resize( l_Search );
for (int i=0;i< beam.size();++i)
beam[i].resize(w_Search);
//初始化柱
//设定系数//生成模板
double gap = 2.0/ (n-1);
std::vector< double > mask(l_Search);
for (int i=0;i< mask.size();++i)
mask[i] = -1 + i*gap ;
//std::cout<< " mask[i]:" << mask[i] ;
//2.生成角度图像
//在射线方向上寻找//方法不是太好，但是没有寻找到简单有效的方法
for ( int y=0 ;y< ds.rows;++y )
float* ptr = (float*)( df.data + y * df.step);
for ( int x=0; x< ds.cols; ++x )
//计算角度
//float ag = (float)(cvWish::cosCv(seed,cv::Point2f( x,y ) ) );
//df.at<float>(y ,x) = (float)(cvWish::cosCv(seed,cv::Point2f( x,y ) ) );
*ptr = (float)(cvWish::cosCv(seed,cv::Point2f( x,y ) ) );
++ptr;
//计算差值-导数
for (int y=0 ;y< ds.rows;++y)
float* pf = (float*)( df.data + y * df.step);
float* pg = (float*)( dg.data + y * dg.step);
unsigned char* pd = (unsigned char*)( dd.data + y * dd.step);
for (int x=0;x< ds.cols;++x )
//计算角度
//cvWish::BeamInit(l_Search,w_Search,cv::Point2f( x,y ),df.at<float >(y,x),beam,0);//0表示从中部开始搜索
cvWish::BeamInit(l_Search,w_Search,cv::Point2f( x,y ), *pf ,beam,0);//0表示从中部开始搜索
cvWish::BeamNormal(dg.cols, dg.rows , beam);
#ifdef SHOW_TEMP
int ii =0;
for (;ii<beam.size() ;++ii)
int j=0;
for (;j<beam[ii].size() ;++j)
//canvasSrc.at<cv::Vec3b>(beam[ii][j].first.y ,beam[ii][j].first.x )[0] =255 ;
//canvasSrc.at<cv::Vec3b>(beam[ii][j].first.y ,beam[ii][j].first.x )[1] =0 ;
//canvasSrc.at<cv::Vec3b>(beam[ii][j].first.y ,beam[ii][j].first.x )[2] =0 ;
//cv::imshow("edgeEvolution",canvasSrc);
//cv::waitKey(1);
#endif
//dg.at<float >(y,x)= 0;
//for ( int k =0; k< l_Search; ++k )
// dg.at<float >(y,x) += (float)( mask[k]* ds.at<unsigned char>(beam[k][0].first.y,beam[k][0].first.x) );
//
//int s = abs ( ( (int)(dg.at<float >(y,x) ) )%255 ) ;
//dd.at<unsigned char >( y, x ) =(unsigned char) (s);
*pg = 0;
for ( int k =0; k< l_Search; ++k )
*pg += (float)( mask[k]* ds.at<unsigned char>(beam[k][0].first.y,beam[k][0].first.x) );
int s = abs ( ( (int)(*pg ) )%255 ) ;
*pd = (unsigned char) (s);
++pf;
++pg;
++pd;
ds.copyTo(_dst);
return;
辅助代码：
//功能： 初始化任意角度的一个方柱，大小已经确定：l_Search*w_Search
//沿射线方向 寻找 一个柱//默认 参数 0 从中部开始
//参数 1代表 从底部开始；参数 2代表从top开始
void BeamInit(
const int l_Search, const int w_Search,
const cv::Point2f &pc, const float angle,
std::vector<std::vector<std::pair<cv::Point ,float> > > &beam,
const int bottomOrTop )
assert (l_Search%2 >0);//确定是奇数
assert (w_Search%2 >0);
assert ( beam.size() == l_Search);//不改变大小
assert ( beam[0].size() == w_Search);
//往角度方向延长
cv::Point2f ps(0,0);
const float angleVert = angle+ PI_1_2 < PI_4_2? ( angle+ PI_1_2): ( angle+ PI_1_2)- PI_4_2;
cv::Point2f pIdx(0,0);
switch (bottomOrTop)
case 0:
//往底部移动
ps.y = pc.y-(0- sin(angle)*l_Search/2);
ps.x = pc.x-(0- cos(angle)*l_Search/2);
//往左边移动
ps.y -= 0- sin(angleVert)*w_Search/2;
ps.x -= 0- cos(angleVert)*w_Search/2;
//对每个点计算
for (int i=0;i< beam.size();++i )
pIdx.y = ps.y +(0- sin(angle)*i);
pIdx.x = ps.x +(0- cos(angle)*i);
for (int j=0;j< beam[i].size();++j )
beam[i][j].first.y = pIdx.y +(0- sin(angleVert)*j);
beam[i][j].first.x = pIdx.x +(0- cos(angleVert)*j);
//int iwCenter = w_Search>>1;
//int ilCenter = l_Search>>1;
先算十字中间
//for ( int i=0; i< beam.size(); ++i )
//
// beam[i][iwCenter].first.x = 0- cos(angle)*i;;
//
break;
case 1:
//往底部移动//默认底部，因此不需要移动
ps.y = pc.y;//-(0- sin(angle)*l_Search/2);
ps.x = pc.x;//-(0- cos(angle)*l_Search/2);
//往左边移动
ps.y -= 0- sin(angleVert)*w_Search/2;
ps.x -= 0- cos(angleVert)*w_Search/2;
//对每个点计算
for (int i=0;i< beam.size();++i )
pIdx.y = ps.y +(0+ sin(angle)*i);
pIdx.x = ps.x +(0+ cos(angle)*i);
for (int j=0;j< beam[i].size();++j )
beam[i][j].first.y = pIdx.y +(0- sin(angleVert)*j);
beam[i][j].first.x = pIdx.x +(0- cos(angleVert)*j);
break;
case 2:
//往底部移动//默认顶部，因此需要移动到底
ps.y = pc.y -(0- sin(angle)*l_Search);
ps.x = pc.x -(0- cos(angle)*l_Search);
//往左边移动
ps.y -= 0- sin(angleVert)*w_Search/2;
ps.x -= 0- cos(angleVert)*w_Search/2;
//对每个点计算
for (int i=0;i< beam.size();++i )
pIdx.y = ps.y +(0- sin(angle)*i);
pIdx.x = ps.x +(0- cos(angle)*i);
for (int j=0;j< beam[i].size();++j )
beam[i][j].first.y = pIdx.y +(0- sin(angleVert)*j);
beam[i][j].first.x = pIdx.x +(0- cos(angleVert)*j);
break;
default:
//和case 0 相同，默认设定为中间位置
//往底部移动
ps.y = pc.y-(0- sin(angle)*l_Search/2);
ps.x = pc.x-(0- cos(angle)*l_Search/2);
//往左边移动
ps.y -= 0- sin(angleVert)*w_Search/2;
ps.x -= 0- cos(angleVert)*w_Search/2;
//对每个点计算
for (int i=0;i< beam.size();++i )
pIdx.y = ps.y +(0- sin(angle)*i);
pIdx.x = ps.x +(0- cos(angle)*i);
for (int j=0;j< beam[i].size();++j )
beam[i][j].first.y = pIdx.y +(0- sin(angleVert)*j);
beam[i][j].first.x = pIdx.x +(0- cos(angleVert)*j);
break;
return;
//对柱规整，使其不超出边界
void BeamNormal(
const int width ,
const int height ,
std::vector<std::vector<std::pair<cv::Point ,float> > > &beam
)
int w = width -1;
int h = height -1;
//对每个点 s
for ( int i=0;i< beam.size();++i )
for (int j=0;j< beam[i].size();++j )
if (beam[i][j].first.x < 0 )
beam[i][j].first.x = 0;
if (beam[i][j].first.y < 0 )
beam[i][j].first.y = 0;
if (beam[i][j].first.x >w )
beam[i][j].first.x = w;
if (beam[i][j].first.y > h )
beam[i][j].first.y = h;
return;
代码效果：
以上是关于Sobel算子取代：基于特定点方向的canny边缘检测的主要内容，如果未能解决你的问题，请参考以下文章
opencv入门之九Opencv边缘检测：Canny算子，Sobel算子，Laplace算子，Scharr滤波器
常见边缘检测对比（Roberts算子Prewitt算子Sobel算子Laplacian算子Canny算子）
常见边缘检测对比（Roberts算子Prewitt算子Sobel算子Laplacian算子Canny算子）