Towards autonomous adaptation in visual tasks.
Boyán Bonev, Miguel Cazorla · 2006
In this paper we study an appearance-based visual recognition approach. The only fea-tures that we extract from the images are ba-sic filters like some edge and corner detec-tors, and color filters. This idea is biologi-cally inspired. We construct a bank of fil-ters and automatically select a small feature set which is appropriate for the given recogni-tion task. We present interesting results about Feature Selection showing that very small fea-ture sets can yield better classification per-formance than the complete set. We analyze this method's generality and performance in indoor and outdoor environments. 1