Research on Key Technology of Multi-Source Remote Sensing Image Fusion
|School||National University of Defense Science and Technology|
|Course||Information and Communication Engineering|
|Keywords||Image Fusion Image Registration Feature Extraction Multi-source Image|
Multi-source remote-sensing image fusion integrates image information from different aerial remote sensors, and some fusion strategy is adopted to get more complete and essential cognition of the specified target. Nowadays, there are wide applies and great actions of multi-source image fusion in national economy and national defence. So it is of great importance to make the related research.This paper works on the two key points: image registration and fusion algorithm. The major work and some important innovations are as follows:Image registration is the foundation of image fusion. First,the concept,the significance, the common methods and models of registration has studied in detail.Then, a automatic image registration method based on multi-features is proposed. The characteristic of this approach is that the invariant moment shape descriptor is used to establish correspondences between the potentially matched regions detected from the two images, and the improved signature is used to find the Registration Control Points(RCP) as the stage of coarse registration. Then the improved chain-code matching has been performed as the registration refinement. Finally, the registration parameters are computed by using the Least Mean Square. The algorithm proposed in the paper adopts not only the centroid of the matched regions but also the salient points along the contour of the matched regions. And due to the guidance of the region registration, our algorithm is robust and efficient. Experimental results with various kinds of image data have shown the high accuracy of our algorithm in multi-sensor image registration.Then,three levels of image fusion technology: pixel-level, feature-level and decision-level are studied.Basic concepts and principles of image fusion are introduced,and many traditional algorithms are then systemically analyzed.With detail comparison of their characteristics and performances,a series of useful conclusions are given.As for feature-based fusion,this paper proposed an edge extraction method based on wavelet transform and multi-source image feature fusion. First of all, wavelet transform is adopted to extract the multiscale edges information from each image. Then the edges are fused in each scale by local modulus maximum method. Finally, the fused feature of edges is combined using their edges information in different scales. The experimental results with optical, infrared, SAR images show our approach is validity and adaptability.