The images from sensors often contain a lot of noise. Aiming at the limitations of two-dimensional histogram algorithm which commonly used in image noise detecting and separating as well as the features of traditional FCM clustering algorithm, a method of FCM clustering prototype initialization based on mathematical morphology is presented: firstly, the image initial clustering cluster is determined based on the theory of mathematical morphology, and then the cluster is processed through the algorithm of FCM. Theoretical analysis and experiment show that the approach can effectively eliminate the image noise and has certain application value for processing the images with uniform noise background.