Local average fitting active contour model with thresholding for noisy image segmentation

In this paper, an active contours model using neighborhood average fitting with thresholding is proposed for noisy images segmentation. Energy of the proposed model is formulated according to the difference between the local average and global region information. For images corrupted by noise, the neighborhood average method is capable of denoising at the expense of blurring images to some extent. However, problems that appear with average method can be settled by thresholding in this work. Minimization of the energy associated with the active contour model is then implemented in a variational level set framework. Moreover, to eliminate the need for costly re-initialization procedure, a reaction-diffusion method is adopted to regularize the level set function for stability. Experimental results on synthetic and real images validate the effectiveness of the proposed approach.