so, the openCV examples in Mastering OpenCV With Practical Computer Vision Projects are pretty great. The simple kernels I was using aren't nearly as good as the Laplacian filter I've got a sketch-image that I am going to work with for now. I'll come back to tweaking the sketch- making it more scratchy/grainy-what have you. I learned today about scale invariance and contour continuation...
Mat gray;
cv::cvtColor(frame, gray, CV_BGR2GRAY);
const int MEDIAN_BLUR_FILTER_SIZE = 11;
medianBlur(gray, gray, MEDIAN_BLUR_FILTER_SIZE);
Mat edges;
const int LAPLACIAN_FILTER_SIZE = 5;
cv::Laplacian(gray, edges, CV_8U, LAPLACIAN_FILTER_SIZE);
Mat mask;
const int EDGES_THRESHOLD =80;
threshold(edges, mask, EDGES_THRESHOLD, 255, THRESH_BINARY);
// cv::erode(fore,fore,cv::Mat());
//cv::dilate(fore,fore,cv::Mat());
// cv::findContours(fore,contours,CV_RETR_EXTERNAL,CV_CHAIN_APPROX_NONE);
// cv::drawContours(frame,contours,-1,cv::Scalar(0,0,255),2);
drawMat(gray, 0,0);
//cv::erode(edges,edges, cv::Mat());
//cv::dilate(edges,edges, cv::Mat());
cv::bitwise_not(edges, edges);
drawMat(edges, 640,0);
Background subtraction on a flat video feed is next on the list, and seems more confounding than doing it with the Kinect and the ofxKinect library. I was working with it today, I'm grasping the power of the Mat object much better although first derivative and related functions are still confounding me because cv::subtract etc are still not producing what I expect. Eg, right now the foreground and the background mask are blended... need to modify the alpha channel I guess...
