Detection of Interest Points


This demo intends to present several methods (we recently developped one of them). Given an image, it creates a coarse representation when only locii of interest (in our case, by means of local multiresolution extrema of contrast energy) are enhanced. The other methods are referenced as corner detectors. You can also get a filtered image based on area of interest.
You can use this demo whith your image but you must specify the entire address (http://....).


Method ?

For the multiresolution contrast-based methods, you can use a parameter related to the quality of the interest points you want, using the measure between 0 and 1.

For the Harris' detector, you have to specify the coefficient in the quality measure (if you do not know about it, just use the default value).

For the Zheng's detector, you have to specify the coefficient in the quality measure (if you do not know about it, just use the default value).

For the Burt's pattern tree technique, you have to specify the threshold in the Laplacian pyramid (the technique keeps only the values over this threshold) (if you do not know about it, just use the default value).

Do you want the key points to be superimposed on the original image ?


More details on our technique in this report

And for the others:

Another demo on this topic

Warning: The process and file format conversions performed by this system can be rather time consuming... It might take a couple of minutes for the result to be returned.


For comments or suggestions.
Jean-Michel JOLION (jolion@rfv.insa-lyon.fr)
Tel: (+33) 4 72 43 87 59
Fax: (+33) 4 72 43 80 97
Reconnaissance de Formes et Vision
INSA Lyon (Bat. J Verne (17 Av. Capelle))
20, Avenue Albert Einstein
Villeurbanne, 69621 cedex, FRANCE