Nestle: Interest point extraction via nested circles

Abstract

A novel low complexity feature extraction algorithm, only performing by a single comparison per pixel on the average during detection is proposed. While single-scale version of the algorithm remains quite efficient compared against the complexity of the state-of-the-art algorithms, a multi-scale version is also proposed to handle blur and scale changes. The performance tests on the repeatability of these keypoints signify the promising performance of the proposed algorithm to be used in many resource limited computer vision applications due to its efficiency and competitive repeatability performance.