Edge Flow Based Level Set Evolution Approach for Medical Image Segmentation

Abstract

Image segmentation is one of the key tasks in different applications of image processing, computer vision and image analysis. A variety of image segmentation techniques are available in the literature, among which level sets based image segmentation approaches are widely used. Although these techniques perform well in the segmentation of synthetic and real images which have strong object boundaries. However, when it comes to medical images where the object boundaries are weak in terms of intensity and are prone to boundary leakage, these methods give poor segmentation results and lead to inaccurate boundary detection. To remedy this problem, we have proposed a novel level set formulation for medical image segmentation which is based on the integration of edge following algorithm into the level set formulation. We have also introduced a new edge indicator function which incorporates average edge magnitude and direction information. Moreover, an implicit stopping criteria for level set evolution is devised which controls the motion of evolving curve and stop the evolving curve at object boundaries. The proposed method is evaluated using two different medical images datasets of different imaging modalities i.e. Dermoscopy images dataset (PH2) and vital stained chromoendoscopy (CH) images. Experimental results show that the proposed method outperforms the other state-of-the-art approaches that have been considered in this thesis.