Aerial Video Images Registration Based on Optimal Derivative Filters with Scene-Adaptive Corners

Abstract

In many registration problems of an aerial video images, computational complexity and precision is of critical importance. In this paper an new aerial video images registration algorithm based on optimal derivative filters with Scene-Adaptive Corners is proposed. Firstly, the Harris detector based on optimal derivative filters is presented and scene-adaptation is used to control the number of feature points, Then through comparing the euclidean distance of the SURF(speed up robust feature) descriptors defined on the corner neighborhoods, the corresponding matches are established. Lastly, the transformation parameters are estimated using the invariant of five coplanar points, then the most "useful" matching points are used to register the frames. Experiment results illustrate that the proposed algorithm carries out accurate image registration and is robust to large image translation, scaling and rotation.