The common corner detecting methods only consider the local properties such as extremum curvature value,and are performed with single-scale images,Which often leads to missing true corners,and they are apt to be interfered by noise.In order to overcome these drawbacks,a new synthetical corner detecting method combing multi-scale analysis method based on nonlinear complex diffusion and global and local curvature properties is proposed.In this method,multi-scale scales images can be obtained by applying nonlinear complex diffusion which can preserve the edges very well.Then corners are detected according to not only its local curvature properties,but also the relationship between neighboring points in the contours.The experiments show that this method can get rid of noise disturbance,and detect more and accurate corners than original methods,and can be widely used in image registration and segmentation.