bgouthamb
commented
6 years ago In xf::Threshold, the _maxvalue is not taken as an argument, instead, it's value is fixed to 255.
In your SimpleBlobDetector implementation, the intended _maxvalue is 255 and the fixed value is also 255. Hence, it matches with OpenCV result. In cvStartFindContours function, the intended value is 1 which differs from the fixed value 255. Hence, the difference in result.
The xf::Threshold function will be updated soon, to match OpenCV's cv::threshold.
I use xf::Threshold() to Instead opencv cvThreshold and cv::threshold. when in class SimpleBlobDetector:
_void SimpleBlobDetector::detectImpl(const cv::Mat& image, std::vector& keypoints, const cv::Mat&) const
{
...
//params.minThreshold = 200;
for (double thresh = params.minThreshold; thresh < params.maxThreshold; thresh += params.thresholdStep)
{
Mat binarizedImage;
//cv::threshold(grayscaleImage, binarizedImage, thresh, 255, THRESH_BINARY);
xf::Mat<XF_8UC1, HEIGHT, WIDTH, NPIX> imgInput(grayscaleImage.rows, grayscaleImage.cols);
xf::Mat<XF_8UC1, HEIGHT, WIDTH, NPIX> imgOutput(grayscaleImage.rows, grayscaleImage.cols);
Mat accelMat(grayscaleImage.rows, grayscaleImage.cols, grayscaleImage.depth());
imgInput.copyTo(grayscaleImage.data);
threshold_accel(imgInput, imgOutput, (int)thresh, 255, THRESHBINARY);
accelMat.data = imgOutput.copyFrom();
binarizedImage=accelMat.clone();
I can get the same result as I got from opencv 2.4.5. But when I use it at the end of function cvStartFindContours() in contours.cpp
_CV_IMPL CvContourScanner cvStartFindContours( void _img, CvMemStorage storage, int header_size, int mode, int method, CvPoint offset ) { ... / converts all pixels to 0 or 1 / if( CV_MAT_TYPE(mat->type) != CV_32S ) { //cvThreshold( mat, mat, 0, 1, CV_THRESH_BINARY ); xf::Mat<XF_8UC1, HEIGHT, WIDTH, NPIX> imgInput (mat->rows, mat->cols); xf::Mat<XF_8UC1, HEIGHT, WIDTH, NPIX> imgOutput(mat->rows, mat->cols); Mat accelMat((const CvMat *)mat), dst0=accelMat; imgInput.copyTo(accelMat.data); threshold_accel(imgInput, imgOutput, 0, 1, THRESHBINARY); accelMat.data = imgOutput.copyFrom(); accelMat.convertTo(dst0, dst0.depth()); } return scanner; }
It can not get the same result as i got result when I using opencv Did I do something wrong? And how can I fix it?