Abstract: Described in the paper here is a high quality skeletonization algorithm of the line images, which emphasizes acquiring accurate data of joint points. Firstly, the conception of cross section sequence （CSS） is proposed. If a line joining two points which lie on the two sides of a raster curve is approximately perpendicular with the local direction of this curve, this pair of points define a cross section. A group of sequent cross sections compose of a CSS which describes the contour of the straight or curved line segment. According to those contour points which do not belong to CSSes, the positions and the topological structures of joint points can be retrieved by calculation and analysis. Each two CSSes which cross the same joint points can be retrieved by calculation and analysis. Each two CSSes which cross the same joint point and fulfill with the smoothing condition are merged as a part of a raster curve which is called the coorigin line segment （CLS）. Finally, the completed skeleton of the line image can be extracted by calculating the middle lines of CLSes. The experimental results show that the joint point based algorithm can well overcome the defects caused by conventional thinning algorithms, such as the distortion of the skeleton at the joint points, a lot of surplus short lines, etc. This algorithm can preserve not only the completed and precise geometrical information of raster curves, but also the topological relation among the curves.