C2F-AFE: A coarse-to-fine infrared and visible image registration method based on aggregation feature extraction

Abstract

In recent years, infrared and visible image registration has advanced rapidly due to the wide application of infrared and visible sensor vision systems. However, existing methods remain susceptible to nonlinear intensity differences (NID) and scale differences, while commonly suffering from inadequate feature extraction, low repeatability, and inefficient feature utilization. To address these limitations, we propose a coarse-to-fine infrared and visible image registration method based on aggregation feature extraction (C2F-AFE). First, we develop an aggregation feature extraction method based on maximum phase map and weighted moment map to obtain more repeatable feature points. Second, we construct a projection scale space for infrared images to achieve scale invariance. Third, we design a feature descriptor that combines maximum phase features with absolute phase congruency orientation features to effectively address NID. Finally, we present a fine matching method to establish a coarse-to-fine feature matching framework for accurate registration. C2F-AFE not only addresses NID and scale differences but also achieves more reliable matching by extracting more repeatable feature points and enhancing utilization efficiency, thereby improving registration accuracy. Experiments demonstrate that C2F-AFE outperforms existing methods in feature matching and image registration, enabling effective and accurate registration of infrared and visible images.