The Linear Edge Extraction with complicated background in High Resolution SAR Images Based on the DS Evidence Theory

An approach of linear object edges extraction in high resolution SAR images with complicated background is presented based on the DS Evidence Theory.The edge model of linear object is built firstly.Then a modified ROEWA edge detector is proposed to get the edge intensity and edge direction with the eight directional templates and the conic function.Under the edge direction,a one-to -one mapping based on Hough transformation is designed to improve the computational efficiency.Most importantly,the extraction frame based on the edge model is presented with the DS Evidence Theory,and three groups of basic probability assignment function (BPAF) of DS Evidence Theory are constructed with the characteristics of the linear object,such as the sharp edge variation in the edge point neighbor,the collinear characteristic among the edge points,the statistical characteristics in the inner or side region of the linear object present as low and smooth intensity.The Dempster rule is adopt to realize the evidence fusion.Finally experiments of the urban roads extraction are carried out with aerial SAR images,and results are analyzed to validate the performance of the approach proposed in the thesis.