Abstract: A coal-rock interface recognition scheme was proposed which was based on Mean shift algorithm. Firstly, principle of Mean shift algorithm was introduced. The pixel position vector and gray value were concatenated to create a joint spatial-range domain. In order to remove isolated areas and spurious boundaries, a selection method of bandwidth parameters in Mean shift were given, which was applicable to coal-rock image segmentation. Simulation results using synthetic and real coal-rock images show Mean shift algorithm can get coal-rock boundary more accurately than K-means algorithm.