Enhanced correlation coefficient as a refinement of image registration

Abstract

A study of the effectiveness of Enhanced Correlation Coefficient (ECC) on the performance of feature-based image registration approaches is carried out. This investigation determines if ECC improves image registration performance on datasets which test on invariance to scale, rotation and viewpoint change. Five state-of-the-arts methods are considered, namely KAZE, Binary Robust Invariant Scalable Keypoints (BRISK), Oriented FAST and Rotated Brief (ORB), Speeded-Up Robust Features (SURF), and Scale-Invariant Feature Transform (SIFT). Root-mean-squared error of control points is used to evaluate the image registration performance on datasets taken from the Oxford Robotics Database. A global ranking factor is used to rank each method within a dataset. The efficiency of each method is recorded as a guide for selecting a method for a specific application. Results indicate that ECC improves image registration performance in most cases with a small time addition.