Abstract: Aiming at the problem of image noise due to the low surface quality of the workpiece or the presence of burr, an improved method of sub-pixel edge detection based on the neighborhood of the pixel using multiple iterative filtering methods is proposed. The principle and implementation method of this algorithm are described. The method is applied to the measurement of diameter of the gear shaft, and it is found that the relative error of measurement can be controlled within 0.2%, and the positioning accuracy is higher than that of the unimproved algorithm. The experiment proves that the algorithm can give out more smooth and coherent edges from the noise image, thus obtaining accurate sub-pixel edge coordinates.