In order to process multi-threshold image segmentation automatically,the traditional pulse coupled neural networks model is improved.A new algorithm of multi-threshold image segmentation using improved PCNN based on the maximization of mutual information is put forward according to the relationship between original image and segmented image,which is based on the optimization object of maximal of mutual information and a new measurement criterion for determining the number of clusters in an image called difference of mutual information.Theoretical analysis and simulation results indicate that the new method can automatically determine the optimal cyclic iterative times and the optimal number of gray-scale clusters,has a favorable capability to carve up characteristics and maintain the edges,texture and details of images,has higher precision in different image segmentation and can be more adaptability.