In the traditional region growing algorithm, because of the limitation of the growing rule, it is easy to produce over segmentation phenomenon in medical image segmentation. In order to solve the problem of over segmentation, a region growing method based on distance constraint and multiple sub points weighting is proposed to segment the cochlear region in temporal bone MR images. Firstly, a distance constraint is added to the region growing method to solve the over segmentation problem in the original algorithm. Then, the dynamic multi seed weighted growing rule is introduced to reduce the error caused by the unsuitable selection of seed points. After experiments on several groups of data, the average values of Dice coefficient and intersection union ratio of the improved algorithm are 90.13% and 88.59% respectively, and the average values of area overlap error and relative area difference are only 11.64% and 9.89%, which are better than other algorithms.