Image Descriptors / Features and Saliency Maps

This is an implementation of part of Kanan & Cottrell (2010), which extracts dense features from color images which can then be used for object or face classification. In our experiments they perform quite well, generally about as well as SIFT descriptors, if not better, on a variety of popular datasets like Caltech-101.

As a side effect the algorithm also produces saliency maps which can be used to find interest points or to predict eye movements.

See http://www.chriskanan.com/nimble for more information.