Abstract: The authors put forward an automatic segmentation algorithm for tomographic images based on the combination of object detecting and region growing algorithms.The relation graph and correlation degree are firstly defined by the strict rules based on the continuity of the object to be segmented between two sequential images.Then several useful shape parameters of the object in the previous segmented image and relation graph,including the center and the least surrounding boxes,are obtained by object detecting algorithm.Finally effective seed points,which are generated according to the correlation degree,is used to initialize the parameters of region growing algorithm for the latter slice.The experimental results show that the new algorithm can automatically segment serial images,and improve efficiency by a factor of about 50% compared to the algorithm of voxel-based region growing.