object recognition - best approach for template matching of binary (edge) images -
to skimage , opencv gurus, given:
- scene image
- template image
what best approach find cross in scene image ? these output smoothing, , canny filters. now, tried kinds of examples in skimage, , opencv template matching results not satisfactory.
my ideal solution rotation, translation invariant (scale invariant bonus) . there way convert contour points , them registration point cloud ? more accurate ? thought ransac how give inputs ransac?
thanks
my approach solving similar problem create large set of rotated , scaled variations of template image , use opencv's matchtemplate function.
i recommend preprocessing step of filling detected , closed contours (for both template , scene image) white since largely black template image might create false positives in black regions of scene image.


Comments
Post a Comment