An Orientation-Robust Local Feature Descriptor Based on Texture and Phase Congruency for Visible-Infrared Image Matching

This letter presents a novel local feature descriptor, entitled "scale-orientation robust infrared features" (SORIF), specifically designed to be robust to geometric variations, particularly to rotations, while dealing with the nonmonotonic intensity variations between images from the visible (VIS) and infrared (IR) spectra. The method calculates the keypoint's orientation using phase congruency, rotates the window for alignment, and then extracts texture and edge features using Log-Gabor filters. These features are compiled into a single vector, forming a descriptor that effectively handles texture and maintains consistency across different orientations of the image, showcasing its robustness in addressing geometric issues in remote-sensing images. We evaluated the proposed descriptor using four different datasets, extensively used in previous works and composed of images taken from the visible and IR spectra. The experimental results revealed that the proposed descriptor is robust to VIS/IR images' nonmonotonic intensity variations and geometric changes. Moreover, it had a superior matching performance, outperforming some state-of-the-art algorithms.