A modified fast local region based method for image segmentation

Abstract

Among many different methods of image segmentation, local region based algorithms received noticeable attention. In particular, these algorithms can be practically useful for images with intensity inhomogeneity like medical ultrasound images and tumor images. Although local based algorithms, unlike edge based methods, are insensitive to noise and weak edges, presence of severe noise in the images is still a problem. In this study, a modified local region based method is proposed for solving this problem, additionally the time consuming re-initialization process is removed by using the robust and fast Reaction Diffusion method. Conclusively, the final results reveal that the proposed modified algorithm can achieve more appropriate results in terms of accuracy and speed than five well-known local based methods.