A new algorithm of nondestructive, automatic and high-speed inspection is proposed for dark-field image to be used to inspect online optic component damage on high power laser system. This algorithm is based on the theory of clustering in pattern recognition. The position of the damage can be automatically analyzed according to this algorithm from its dark-field image. Then, according to the characteristics of damage image, the two-directional (2D) scanning way can be used to obtain a continuous damage block without leaving out any damage pixel. Theories analysis and experiment show that the algorithm can determine the position of optical component damage and analyze the damage dimensions accurately and automatically.