A novel approach for the registration of weak affine images

A novel image registration scheme is devised in view of the weak affine transformation, which is a kind of similarity transformation with anisotropic scales or affine transformation without shearing. Two robust algorithms are proposed to retrieve the registration parameters from the error-prone initial correspondences based on the fast least trimmed squares (Fast-LTS) and the random sample consensus (RANSAC). In terms of several criteria, the algorithms are evaluated on three carefully selected datasets from different sensors and the experimental results demonstrate that the proposed scheme and algorithms perform robustly and accurately. Our findings also indicate that the Fast-LTS-based algorithm is more stable and appropriate for image registration than the RANSAC-based algorithm although the speed is slower.

► An image registration scheme for WAT is devised in view of anisotropic scales, rotation, and translation.
► Two algorithms based on Fast-LTS and RANSAC are proposed to retrieve the parameters from the contaminated correspondences.
► Experiments on datasets from different sensors show the validity and robustness of the proposed scheme and algorithms.
► Estimation algorithm based on Fast-LTS is more stable and appropriate for image registration than that based on RANSAC.
► WAT can model the AT warped images with tiny shearing and the real-life ST warped images with unavoidable measuring error.