We would need to find a bounding curve outlining the image with OpenCV findContours and reduce it to a certain number of points with approxPolyDP.
Then for each of the points around the boundary can be considered a keypoint. For each, the descriptor is given by a weighted sum of the angles between the pair of lines which connect this point to all the others in the boundary at a certain number of points both forward and backward along the curve. The weight is inversely proportional to the perimeter distance between the two points for which the angle was computed.
The description should possibly be coupled with a different description of the same points (SIFT is used in the paper, but we can't use that).
The BAS (beam angle statistics) method described in this paper: http://www.sciencedirect.com/science/article/pii/S0167865503000023
and used in this paper as a description. http://web.engr.oregonstate.edu/~tgd/publications/WACV2011-stacked-spatial-pyramid.pdf
We would need to find a bounding curve outlining the image with OpenCV
findContoursand reduce it to a certain number of points withapproxPolyDP.Then for each of the points around the boundary can be considered a keypoint. For each, the descriptor is given by a weighted sum of the angles between the pair of lines which connect this point to all the others in the boundary at a certain number of points both forward and backward along the curve. The weight is inversely proportional to the perimeter distance between the two points for which the angle was computed.
The description should possibly be coupled with a different description of the same points (SIFT is used in the paper, but we can't use that).