Research On Multimodal Remote Sensing Image Registration Based On Cross-Cumulative Residual Entropy

Image registration, which is the foundation of image change detection, imagefusion, target recognition, image stitching and other image processing technology, hasa wide range of applications, such as agricultural monitoring, environmentalprotection, resource exploration, missile guidance, and so on. As the increasing ofremote sensing images of different data types, the remote sensing image registrationtechniques developed from the single-mode image registration to multi-mode imageregistration. The remote sensing image registration research status is summarized inthis study. Based on the good sparse representation of NSCT transform decompositionand the effectiveness of the registration process of cross-cumulative residual entropy(CCRE), a multi-mode remote sensing image registration algorithm is purposed.Detailed remote sensing image registration techniques are discussed based on thisalgorithm.In this paper, the theory of remote sensing image registration module-geometrictransformation and common image interpolation algorithm is analyzed. Based onthese theory, radiation transformation and bilinear interpolation algorithm is utilizedfor image registration. This paper makes summarize on the development ofmulti-scale decomposition and analyze the NSCT theory in detail. The superiority ofNSCT is its translation invariance, multi-scale and multi-directional.Cross-cumulative residual entropy, which can be adapted to different brightness andcontrast of the image, has been widely used in practical applications as an imageregistration similarity measure. It has good result when processing with a largeoverlap region of two images, as well as two images with a large range oftransformation parameters. It can have better treatment effect than a similar measurein the mutual information image registration on superior convergence faster, lesscomputation and higher precision. In the paper we analyze the advantages of Newton’s method is selected as the image registration process optimization algorithm,and give the optimization process of Newton’s method for CCRE similarity measure.In addition, this paper introduces the characteristics information of two multi-moderemote sensing images used in the experiment. The experiment result shows that thefast and efficient multi-mode remote sensing image registration algorithm weproposed can quickly search the global optimal solution and has high registrationaccuracy. The last section of the article is a summary of knowledge of the structuralframework and the prospect of the next step in image registration based on themulti-mode remote sensing image registration algorithm proposed in this paper.