Perceptual Contrast Based Image Registration and Fusion using Moving Least Squares and Multi Component Analysis

Abstract

This research focuses on exploring an effective and efficient method for registration and fusion of medical images. In this work, we have analyzed the applicability of Moving Least Squares (MLS) Transformation with the as-rigid-as-possible constraint for registration of medical images that are non-rigid in nature. The MLS being a point based method requires the selection of control points from the source and target images. Therefore, we have also carried out the analysis on the consequence of automating the control point selection process using feature extraction algorithms like Harris Corner, Min-Eigen, Speeded Up Robust Features (SURF) and Canny Edge pixels and the effect that the number and position of control point have on the registration process. Since the end users are medical practitioners, who prefer to have an interactive system, where the control points need to be selected based on the diagnostic requirements manual selection of control points is also included. The selected points from the source image and the target image need not necessarily be the corresponding points. These points have to be matched to identify the corresponding points between the two images. For this, a new structural method for control point matching method is proposed. For structural matching the control points are represented using a graph structure and the structural properties like the degree of the vertex, length of edges and the angle between edges are used for finding the corresponding points in the source image and the target image. This method is found to be efficient for both mono-modal and multi-modal image registrations, as the topological property represented by the control points are exploited instead of the traditional intensity feature. The accuracy of the registration is computed using the standard Target Registration Error (TRE) Measure. The registered images are then fused to get a single image that contains the perceptual and structural information from the input images. ..