This paper discusses the method of straight-line extraction

which is one of the open problems in computer vision. Based on the analysis of several multi-scale approaches for edge detection

a new idea which straight lines are considered as a kind of feature structures in scale space is developed. Then straight-line extracting is converted into the procedure of analyzing the features in scale space. After the characteristics of the line structures in scale space are analyzed

a robust approach for straight-line extraction is presented. Experimental results on a synthetic image and several natural scene images indict new algorithm's effectiveness.