Retina enhanced descriptors for spatio-temporal concept detection.

Date de publication : Oct 29, 2012 5:29:49 PM

Great news, the paper "Retina Enhanced SURF Descriptors for Spatio-Temporal Concept Detection" (DOI: 10.1007/s11042-012-1280-0) has been accepted for publication at MTAP !

In this paper is described how to use the retina as a preprocessing tool for enhanced visual features extraction. An automatic transient blob detector is also proposed. It takes advantage of the transient retina channel (magnocellular channel) to detect blobs related to &. low level saliency (detailledeye catching areas when opening your eyes) and moving areas. We propose to use such blobs to perform Bag Of Words extraction... and it gives interesting results.

See on the following figure a view of these transient blobs. It shows the 3D volume of a video shot. during the first frames, blobs are related to eye catching areas, then as time goes, retina gets stable and reacts on events (motion and movie cuts). More details in the paper, check the Publications pages.