A robust affine parameter estimation using acquainted non-convexity multiresolution based optimisation for image registration

Abstract

Parameter estimation technique applied to image registration is found useful in obtaining reliable fusion of same object's images taken from different modalities, into single image with strong features. In this paper, the proposals are two-fold. A refined approach of non-local means filter for the purpose of preprocessing in image registration is proposed. Then we propose a new optimisation strategy for gradient-based image registration. Usually, the minimisation in image registration technique can be done by least squares in a quadratic way. However, this will be sensitive to the presence of outliers. Therefore, image registration technique calls for the methods that are robust enough to withstand the influence of outliers. Progressively, some robust estimation techniques demanding non-quadratic and non-convex potentials adopted from statistical literature have been used for solving these. Addressing the optimisation of error function in an efficient framework for finding the global optimal solution, the optimisation can begin with the convex M-estimator at the coarser level and gradually acquaint non-convex M-estimators when the iteration reaches finer level of multiresolution pyramid. Comparison has been made to find the performance results of the proposed method. Results show significant improvement with respect to the standard estimation techniques.