Extension of the retina preprocessing concept for video indexing.

Date de publication : Apr 20, 2013 6:45:11 AM

The paper : S. Strat, A. Benoit, P.Lambert, "Retina enhanced SIFT descriptors for video indexing", CBMI2013, Veszprém, Hungary, 2013 has been accepted for publication. In previous papers, Retinal preprocessing on video prior to image description proved to improve SURF descriptor performances.

in this paper, we show that a similar enhancement can be obtained with SIFT.

Retina behaviors and features sampling where modified and the enhancement is still observable proving the interest of the approach.

Finally, a new descriptor is proposed. It relies on the description with SIFT of the two well know pathways of the retina (parvocellular: detailled central vision and magnocellular peripheral vision sensitive to transient events (motion&changes). This new descriptor presents new behaviors and is able to describe static and motion patterns. It is complementary with the previous approaches for video indexing.