Corner Detecting Based on Nonlinear Complex Diffusion Process

This paper proposes a technique for detecting corners of medical images based on complex diffusion process. A corner detecting method based on global and local curvature properties has been traditionally used in extracting feature points in images. This kind method is ineffective for objects which have multiple-size features since the curvature measures are only based on single scale regions. Corner detection could perform better at multiple scales. The usually used multiple scale space function which is generated by convolving a 2-D image with a varying-scale Gaussian smoothing function often blurs the edges of the image and causes important information loss. Anisotropic diffusion methods can provide a hierarchical representation of images in scales ranging from coarse to fine. Nonlinear diffusion process of a partial differential equation (PDE) can make edges preserved and has been widely used in image enhancement and denoising. With the PDE method, we extend the analysis from the real axis to the complex domain. An important observation is that in complex diffusion, the imaginary part of the image can serve as an edge detector (smoothed second derivative scaled by time) when the complex diffusion coefficient approaches the real axis. A synthesis corner detecting method based on complex diffusion, and global and local curvature properties is proposed, and the results show that more corners at different real and imaginary part of the image can be detected, which is important for segmentation and registration of images.