关键词: 图像配准, 红外图像, 非线性变换, 形状上下文

Abstract:

Image registration is a key problem for infrared and visible image fusion. In real applications, non-planar scenes and differences between infrared and visible cameras can increase the difficult of multimodal image registration. Aim at this problem, Adaptive Polynomial Mixture Transformation（APMT） is proposed for image registration, which is able to describe the non-linear regularity of deformation between infrared and visible images to be aligned. In addition, Gaussian Weighted Shape Context（GWSC） is developed to extract putative matches from multimodal images. For achieving global image registration, the optimal APMT model is then estimated from the putative matches between infrared and visible images by a strategy of subsection optimization. The qualitative and quantitative comparisons demonstrate that GWSC-APMT performs well on image registration, being superior to several competitive approaches on registration accuracy and speed.

Key words: image registration, infrared image, non-linear transformation, shape context