基于自适应混合多项式变换的图像配准

Image Registration via Adaptive Polynomial Mixture Transformation

图像配准是红外与可见光图像融合的关键问题。在实际应用中,场景景深的多变性与红外、可见光相机之间的差异性都会增加多模图像配准的难度。为应对上述困难,提出了一种用于图像配准的自适应混合多项式变换(Adaptive Polynomial Mixture Transformation,APMT),该模型可以准确地描述待配准红外与可见光图像之间形变的全局非线性规律。针对形状上下文特征的缺陷进行改进,设计了高斯加权形状上下文(Gaussian Weighted Shape Context,GWSC)特征,用于从多模图像中提取匹配点集。利用分段优化策略从匹配点集中估计出最优的APMT模型参数,实现全局图像配准。定性与定量实验表明:与同类方法相比,提出的方法(GWSC-APMT)在配准精度与效率方面都有良好的表现。

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.