Composite Feature Based Heterogenous Images Registartion Algorithm Research

Image registration is an important research field in image processing,and it is also an essential foundation for image splicing,fusion,and reconstruction.Along with the de-velopment of research in computer vision and image processing,single sensor are hardly satisfies demand of applications.Heterogeneous images processing are propitious to uti-lize information for application,especially in the fields of medical imaging,militant de-tection and remote sense.Due to the effects of sensory difference,optical principle and scale transformation,heterogeneous images often exists large differences in gradation and are difficult to recover structural relationships.Compared with homologous images,heterogeneous images registration techniques still face many challenges.Because of the difference in gray-scale distribution in heterogeneous images,grayscale-based registration method cannot apply to heterogeneous images.This paper discusses heterogeneous image registration algorithm based on composite features.The algorithm mainly involves the composite features extraction and matching under help from cluster of depth information.We briefly introduce basic theory and method image registration,especially feature-based registration method.Against the problems in heterogeneous im-age registration,SIFT feature point detection and Canny edge detection are used to extract and register the features,against the problems in heterogeneous image registration.In-stead of utilizing difference in gray-level,a iterative Gaussian mixture model are proposed for registration task which is solved with discrete coherent point.For the results of point set registration,a mechanism is proposed for improving performance of features based on depth information.Because of the unreliability of detail information between heterogeneous image,the matching optimization part utilizes robust depth information.By binocular imaging principle,the depth map of the reference image is calculated.a reliable feature area with relevance is extracted from the depth image.Then matching the highly relevant area through segmentation and automatic choose in depth image.Finally,apply the registering method to composite features in target area for solving the deformation between image pairs.Experimental results show that the proposed algorithm performance better for the heterogeneous images registration in both subjective and objective measurement.This demonstrates proposed method can register heterogeneous images successfully and achieves the goal of this research.