Research On Edge-based Matching Method Of Remote Sensing Images

Image matching has been widely used to automatically determine the counterpart features to replace manual stereoscopic observation in digital photogrammetry,which has greatly promoted the development of photogrammetric automation.The edge feature is a basic feature of the constitution of object description,it involves rich information of the object,and relatively stable.Thus,edge-based matching has a big advantage in stability comparing with the conventional gray matching and point-based matching,which is particularly apparent while concerning the matching between multi-sensors images.Research in this paper is mainly focused on edge detection,edge tracing,post processing and edge matching in the procedure of the edge-based matching of remote sensing images.The main work and innovation are as follow:1.This article makes an introduction of the research background and significance.The status quo of edge detection and the matching of the remote sensing images are summarized,conclusion of characteristics of variety matching methods are also involved.2.A universal sub-pixel edge detection algorithm is proposed based on extremal gradient.Extremal gradient is disintegrated into positive and negative gradients that are solved respectively in eight directions.Then,initial edge composed of two types of pixels with local gray level maximum increase and decrease can be obtained.Finally,sub-pixel orientation fitting models are built for different types of edges separately according to the characteristic of initial edges.Contrast experiments have been realized to verify its performance based on simulative and real images.The results indicate that the proposed algorithm has better applicability of different types of edges and higher precision including corner point than traditional algorithms.3.A new edge tracing and post-processing method is proposed,with the purpose of solving the extraction of free-form sub-pixel edge.While tracing the edges,use the connected edge blocks as the basic unit,and take full account of the existence of multiple tracing direction.In order to acquire boundaries as continuous as possible,use a strategy of double length-angle control to extend and connect edges.For noisy edge curves,a bilateral filtering method is used to denoise.The results indicate that the proposed method could extract continuous and smooth free-form edge curves effectively.Therefore,this method offers foundation for edge-based image registration.4.A matching method with a hierarchical strategy based on edge features is proposed.Features like closed edge segments,intersection points and angular points of edges are extracted from continuous edges and regarded as conjugate entity in the rough matching.Firstly,invariant moments of closed edge segment and angle between the branches of intersections and corners are used to determine the conjugate features.Then the combination between every two conjugate features pairs can be used to determine a set of parameters of similarity transformation.Multiple restrictions are used to gradually eliminate combinations which contain mismatch features.After determine the optimal transformation parameters,corresponding features and the approximate value for fine matching transform model can be determined too.To improve the accuracy,low-to-high sampling rate of sub-pixel edge points are used in the precision matching with ICP method.K-D tree search strategy is used to determine the proximal points for a high computational efficiency.Experimental results show that the coarse matching has stable performance and can provide relatively accurate initial value for fine matching,fine matching can reach sub-pixel matching accuracy.