Yes, because current state of the art is developed for textured objects and objects with a lot of details. When we were developing a system that should recognise simple blocks from a shape sorter puzzle we found that none of the existing features were sufficient.
|
|
|
|
| Input image | SIFT output | MSER output | CCF output (Our new feature) |
| Frames either unstable or too large. | Only two frames detected. | Four stable frames and one spurious. |
The colour contour frame (CCF) algorithm was published at the CRV'06 conference.