[Blog Machine vision] maximum contour and projection

maximum contour and projection    The most recent use of the maximum contour and projection operations. In retrospect, these two algorithms are indeed very common basic algorithms. summarized here and extracted. Maximum profile: The premise is that the image is processed by grayscale and threshold, and the results of canny can be processed directly, and in some cases it needs to be converted in color domain beforehand. The final result, should be a contour, of course, can be used to deal with a certain method to get the external rectangle. Looking for the biggest silhouette
VECTOR<CV::P oint> findbigestcontour (Mat src)
{
intIMAX = 0; Number representing the maximum contour
intImaxcontour =-1; Represents the size of the maximum contour
STD::VECTOR<STD::VECTOR<CV::P oint>>contours;
Findcontours (src,contours,cv_retr_list,cv_chain_approx_simple);
for(intI=0;i<contours.size (); i++)
{
intItmp = Contourarea (Contours[i]);//The contour size is used here
if(Imaxcontour < Itmp)
{
IMAX = i;
Imaxcontour = itmp;
}
}
returnContours[imax];
}call method to get the external matrixRect boundrect = Boundingrect (Mat (Findbigestcontour (canny))); projection processing :      the premise and the maximum contour are very similar here. The main concern of projection is to obtain the ROI in the original image by projecting the image, or to obtain the quantitative results of how many down waveforms are in the waveform.
vector<int> Vectorv; Transverse circulation
vector<int> Vectorh; Longitudinal cycle
vector<int> vupper;
vector<int> Vdown;
vector<int> hupper;
vector<int> hdower;

Do the transverse loop
for(inti=0;i<ostu.cols;i++)
{
Mat data = Ostu.col (i);
intItmp = Countnonzero (data);
Vectorv.push_back (ITMP);
}
The upper waveform is vupper and the lower waveform is Vdown
for(intI=1;i<vectorv.size (); i++)
{
if(Vectorv[i-1] = = 0 && vectorv[i]>0)
{
Vupper.push_back (i);
}
if(vectorv[i-1]>0 && vectorv[i] = = 0)
{
Vdown.push_back (i);
}
}
To do a longitudinal cycle, this is often handled by a picture of the result of a transverse loop
for(intj=0;j<ostu.rows;j++)
{
Mat data = Roitmp.row (j);
intItmp = Countnonzero (data);
Vectorh.push_back (ITMP);
}
for(intJ=0;j<vectorh.size () -1;j++)
{
if(vectorh[j]>0 && vectorh[j+1] = = 0)
{
Hdower.push_back (j);
}
if(Vectorh[j] = = 0 && vectorh[j+1]>0)
{
Hupper.push_back (j);
}
}

Due to the processing of the upper waveform and the lower waveform may be problematic, need to do some processing
One of the treatments here is to put forward the short blank areas
for(intJ=0;j{
To draw an area between the blanks
intiwidth = hupper[j+1]-hdower[j];
if(Iwidth > 10)
{
Iresult = Iresult +1;
}
}


From for notes (Wiz)

[Blog Machine vision] maximum contour and projection