A NOVEL IMAGE INTEREST POINTS DETECTOR BASED ON HARRIS METHOD USING ATOMIC FUNCTIONS

Abstract

Most computer vision applications base their operation on the interest point detection. Therefore accurate interest point detection is essential for them. The detected interest points must be invariant to rotation, zoom, blur, illumination, changes of viewpoints, etc. This paper presents a novel image interest point detector, called Harris Atomic detector, which is based on the conventional Harris detector, but the Gaussian Function used in was replaced by an Atomic Function (AF) to improve the detection accuracy. In order to know which Atomic Function provides better results under different imaging conditions, a performance evaluation was done using several AFs. The main criterion used for the comparison was the invariance, in position and number, among the interest points detected in the first and the subsequent frames of a known image sequence. Evaluation results show that, although in general any atomic function improves the performance of conventional Harris detector, the up(x) AF provides best performance, with an improvement of about 10% on the interest point detection accuracy under rotation, blur, zoom and view point changes.