In order to solve the problems of high precision contour measurement with arbitrary shape, a high accurate vision algorithm is proposed based on shape context descriptor and iterative closest point (ICP). Firstly, the sub-pixel edge of workpiece is extracted by using edge extraction algorithm based on local area, then the noise is filtered out and contour is completed. Secondly, based on coarse-to-fine matching strategy, the coarse matching is performed by using shape context descriptor and the fine matching is performed by using iterative closest point for obtaining the sub-pixel accuracy contour matching. Finally, the neighbor-method is proposed to calculate the contour errors. The experimental results show that the detection accuracy of the proposed algorithm can reach to 0.5 pixel, which can meet the needs of practical application.