In order to solve the problem that the traditional image segmentation algorithm is difficult to segment the dam surface crack images with serious noise pollution, a dam surface crack recognition and segmentation algorithm based on adaptive region growing and local K-Means clustering was proposed. Firstly, bilateral filtering to initially reduce noise in crack greyscale images is used, then the adaptive region growing algorithm to obtain the rough segmentation image of cracks is applied. Secondly, the isolated noises like points and groups are removed by morphological eroding and largest connected domain extracting. Finally, the precise segmentation image of cracks is obtained by local K-Means clustering algorithm. The proposed algorithm, Otsu threshold algorithm and other three algorithms are used to segment three crack images which have stain noise, concrete surface burr noise, block noise and strip noise respectively. The results show that the completion index and accuracy index of the segmentation results obtained by the proposed algorithm are above 0.95, which is better than the other three algorithms. The proposed algorithm has good anti-noise performance and strong adaptability, which can realize the accurate identification and segmentation of dam surface cracks.