An evaluation of local interest regions for non-rigid object class recognition

► EdgeLap, SURF, HarAff, HesAff, kAS, FAST, IBR, PCBR, HarLap, HesLap, Salient, MSER, and DoG are evaluated. ► EdgeLap interest region detector reaches to the highest success-rate. ► A new discriminancy measure based on random region success-rate is introduced. ► SURF and DoG interest region detectors have not only good success-rates but also good discriminancy values. ► ExpRand gets success-rates as high as the best detector, and is more discriminant than most detectors.