Noisy Image Segmentation Based on a Level Set Evolution

In this paper, we propose a new hybrid model for active contour image segmentation, which is able to segment non-uniform noisy images efficiently. The model is a combination between the classical active contour based on the image gradient and the mean curvature moving technique. The efficiency is achieved by de-noising the image using log-Gabor filter then using a hybrid model to segment the noise free image. The proposed model has three main advantages over the pervious models and other traditional segmentation techniques. First, a significantly larger time step can be used for numerically solving the evolution PDE, and therefore speed up the curve evolution. Second, the model can be used to segment the image in the presence of high or low noise. Third, the proposed model can be used for segmenting a single object or grouping multi-objects in the image. We will present various experimental results on natural and synthetic images which demonstrate the power of the proposed method for segmentation of noisy image.