Study On Line Reconstruction Method For High Resolution Optical Remote Sensing Image

As a kind of typical geometric optical features of remote sensing image,straight line is an important clue of visual perception and basic basis of image interpretation,which provides a strong support for the basic structure of the graphic recognition in the perception mechanism of human visual guide.But affected by the complex background and occlusion,jagged edges and other factors,linear fracture problem often occurs in the process of line-extraction of high resolution satellite remote sensing image,and the traditional line extraction method based on spectral characteristics cannot solve this problem.In view of this,this paper puts forward a new method of line reconstruction:Firstly,This paper uses improved Canny operator to extract images edge in the phase of edge detection,and puts forward a new refining templates based on classic ZS algorithm,refining the treatment effect and connects gap between the edge pixel group using heuristic methods to improve the integrity and continuity of the edge pixel;Next in the phase of straight line extracting,defining the new chain code tracking starting point,conducting chain code tracking of line edge,tracking the annular edge,analyzing edge points of eight neighborhood inside and outside in the process of tracking,tracking the crossing after the completion of the edge compensation,and according to the chain code tracking order,validating effectiveness of linear dynamic analyst to complete least squares line fitting;Then in the phase of straight line reconstructing,choosing the initial line according to length priority principle to establish a rectangular search area.In the scope of the search rectangular,it sets up geometry,spectral information constraint model to identify line fragment.The initial long line and short line are used to construct linear dynamic reconstruction model.Finally,through the enormous of experiments and analyzing,in the process of reconstruction,the geometric and spectral information constraint model and dynamic reconstruction model figure out it better than other algorithms in the aspect of accuracy,length and the efficiency of the algorithm.