Skip to Main Content
Extracting distinctive scale invariant features from images of the same scene or object is very important in many computer vision applications, and there has been significant research into the scale invariant feature detectors and descriptors. Some of these methods have emphasized on computational speed and accuracy, so that they can enable lots of real-time applications with reduced computational requirements and better performance. The purpose of this paper is to introduce a new scale invariant octagonal center-surround detector, named OCT, and give a modified SURF descriptor named I-SURF descriptor. OCT detector computes responses at every pixel and all scale, and could be implemented efficiently by using integral image and slanted integral image for image convolutions. I-SURF modifies the SURF descriptor by considering the boundary effect of the adjacent subregions, and introduces index vector to speed up matching. The evaluation system provided by Mikolajczyk is applied to OCT and I-SURF. Experiments of the repeatability score, matching accuracy and timing proved that our detector and descriptor have better performance than SURF and SIFT.