Partial Differential Equation Model For Noisy Image Segmentation

Image segmentation and edge detection are the fundamental problems in image processing and computer vision. The goal is to partition a given image into series of meaningful regions and extract the most interesting objects from an image, which makes the analysis easier. So far, a lot of good algorithms including active contour models have been widely applied to image segmentation.In recent years, the active contour models based on level set method, due to its simple implementation and better segmentation effect, have become a hot research topic in the field of image processing. Active contour models can be divided into two categories: edge-based models and region-based models. The region-based models have more advantages than the edge-based models, such as the more accuracy to extract weak and fuzzy edges and robust to the size and location of initial contours. Moreover, they have more strong robustness to the noise.This dissertation discusses level set method and several classical region-based active contour models, such as C-V model, RSF model and Zhang model. Based on the analysis on causes of sensitivity to noise for these models and curve evolution theory and level set method, we propose a partial differential equation for noisy image segmentation. We first construct a new speed function by using local and global statistics information, and then derive an evolution partial differential equation for level set function. Experimental results show that the proposed model has desirable performances for synthetic and real images with high level noise and can well handle images with weak and fuzzy edges. In addition, this model is very robust to initialization of contours. It also can process some skin lesion images and so has a certain application prospect. In addition, compared with Zhang model and C-V model, the proposed model has more advantages in terms of convergence speed and segmentation effect.