Deformed contour segment matching for multi-source images

Abstract

Robust and accurate multi-source matching is a difficult task due to significant nonlinear radiometric differences, background clutter, and geometric deformation in corresponding regions. Motivated by these existing problems, a discriminating yet robust combined descriptor for multi-source image matching, called deformed contour segment similarity (DCSS), is proposed in this work. First, the proposed DCSS, which is constructed by histogram of the combined contour features rather than the commonly used corner point and gradient, presents the accurate correspondence between image pairs and improves the descriptive ability to radiometric differences. Second, the deformed curve is presented via a finite-dimensional matrix Lie group to determine the similarity metric with an explicit geodesic solution. The geodesic distance, which indicates the nearest distance between curves in fluid space, is defined as the weight coefficient of the constructed histogram to enhance the robustness of the descriptor. The proposed algorithm utilizes the holistic contour information for the scoring and ranking of the shape similarity hypothesis, which can effectively reduce the influence of partially missing contours. Finally, a precise bilateral matching rule is used to perform the matching between the corresponding contour segments. Some experiments are carried out on various infrared-visible image data sets. The results demonstrate that the proposed DCSS achieves more robust and accurate matching performance than many popular multi-source image matching methods.