To solve the problem of heart cavity over-segmentation in image processing, a new algorithm is proposed. It combines a new multi-threshold segmentation based on histogram with marker-based watershed segmentation algorithm. Because CT heart cavity region gray values are very similar and the heart tissues are connected with each other, it can't get the target area with the traditional threshold segmentation. The appropriate threshold values are selected according to the corresponding histogram of the CT images in new multi-threshold algorithm, and the segmentation results can be used as input images of marker-based watershed algorithm, eventually the cavity area is constructed via the three-dimension reconstruction. Compared with other methods, the segmentation result is more accurate, and the boundary is clear because of the use of multi-threshold based on histogram. It effectively reduces the over segmentation if only the original watershed algorithm is used.