Novel scale-invariant keypoint detector

We propose a novel method for detecting scale-invariant keypoints. Our method first computes multiscale representations of an image for several different measures by scale-space filtering. Then we localize feature points in the image plane and determine the characteristic scales for these points using a saliency metric. Finally, some highly salient points are selected as keypoints. These points are invariant to scale and rotation. Experimental results show very satisfactory performance of the detector. (c) 2008 Society of Photo-Optical Instrumentation Engineers.