In this paper，the Mumford-Shah model is used for the recognition and classification of satellite cloud imagry.Considering the characteristics of satellite cloud imagery，an improved Mumford-Shah model based on the kernel gray of object is presented.Using this model，high cloud and middle cloud and low cloud is classified from the one-channel cloud image.For the multi-channel image，the vector-valued image segmentation method using Mumford-Shah model based on the kernel gray of object is presented.The experimental results show that our method can get the precise position of middle cloud and low cloud in the infrared channel and visible-light channel.