Basically I need to develop an object recognition system. I have already used with success matching with SIFT or ORB.
Now I would like to use in the same system something else to improve SIFT/ORB, for example shape matching after a canny edge extractor (or whatever)
Can you suggest me a way to proceed?
I was thinking to always start
- with corner detection (like SIFT,ORB)
- then compute ORB descriptors for the corners (like standard ORB)
- now extract some more information around a patch centered in the corner (contour infroamtion?)
Any Hint?