In order to improve accuracy of image Segmentation, a new merging method based on Bayesian classifier is proposed for the medical image Segmentation. There are many particles such as red blood cells, white blood cells, pipe type cells, epitheliums and the crystallizations in urinary sediment images. Segmentation the various elements among the particles is very important to medical decision. Existing plenty of background noises in images, so preprocessing is needed to eliminate those noises before segmentation. Preprocessing adopts the mathematical morphology methods to carry out edge pick-up, gradient graph double value, corrosion and expansion. Then the biggest posterior probability method is used for combination of incomplete object entities during image segmentation. In the end, Bayesian classifier is used for classifying of particles. Experiment shows that the new method is efficient for image segmentation of urinary sediments.