Abstract: The grainplay a decisive role in the tensile strength, toughness and other mechanical properties of the metal. In order to quantitatively evaluate the grains in the metal, the grain boundaries must be determined in the metallographic image. Aiming at the phenomenon of grain boundary blur or fracture, precipitated phase and interference of polished particles in high-power micrographs of high-temperature alloys, a method of grain boundary extraction proposed. mage contrast is enhanced by Laplacian sharpening and histogram equalization, and then segmented by Otsu method. The obtained grain boundary map is used to remove the maximum connected domain and the independent points are removed. Finally, the smoothing operation is completed to obtain some complete boundaries and the characteristic parameters of the grains are calculated. The metallographic images of GH4149, GH706 and GH738 are processed. The results show that the algorithm is close to the effect of traditional software extraction and subsequent processing, eliminating the need for repetitive work and reliability.