ORB-PC feature-based image registration

Abstract

In remote sensing, intensity variation and rotational distortion are unavoidable issues for multimodal image registration. To address the intensity variation issue, Radiation-Invariant Feature Transform (RIFT) feature-based image registration method was developed. The RIFT produced good results for horizontal and vertical transformation in multimodal images. However, the extracted features in RIFT produce some feature matching errors due to the rotational distortion and hence reduces the registration accuracy. Oriented FAST and Rotated BRIEF (ORB) can produce better rotation invariant features since it was basically designed to deal with the mentioned issue. However, it is not suitable for solving the intensity variation problem. Therefore, in this research work, we integrated the Phase Congruence (PC) to ORB as PC is robust to illumination and intensity variations while the ORB is robust to rotational distortion. Hence the combination of both methods has been utilized to overcome all the mentioned issues simultaneously in order to enhance the accuracy of the image registration. Experiments were conducted in order to validate the superiority of the proposed method over recently introduced RIFT method in terms of accuracy.