Oriented filters for object recognition: an empirical study

Local descriptors are increasingly used for the task of object recognition because of their perceived robustness with respect to occlusions and to global geometrical deformations. Our performance criterion for a local descriptor is based on the tradeoff between selectivity and invariance. In this paper, we evaluate several local descriptors with respect to selectivity and invariance. The descriptors that we evaluated are Gaussian derivatives up to the third order, gray image patches, and Laplacian-based descriptors with either three scales or one scale filters. We compare selectivity and invariance to several affine changes. The overall results indicate a good performance by the descriptor based on a set of oriented Gaussian filters. Finally, we discuss briefly an object detection system based on the Gaussian descriptor that we have implemented: preliminary results confirm robust performance in cluttered scenes in the presence of partial occlusions.