A robust multi-level scene matching algorithm for infrared and visible light image

To tackle the problem that image matching methods based on feature have great dependence on feature's quality and have poor adaptability to image noise and local distortion for multi-modal images, this article starts from enhancing the representation ability of images. Then, phase congruency transformation with local intensity and contrast invariance is used to represent the infrared images and visible light images to increase the robustness of scene matching. Based on this, the ring projection transformation is first adopted to get these coarse matching positions. Then, the cross-correlation reconstruction function is inferred using Zernike moment to remove these wrong matching positions. Finally, curve fitting method is used to get sub-pixel localization and realize image accuracy and robust matching. Experimental results demonstrate the effectiveness of this proposed algorithm.