A Level Set Method for Natural Image Segmentation by Texture and High Order Edge-Detector

Active contour model has been a widely used methodology in image segmentation. However, due to the texture complexity of natural images, it unavoidably faces many difficulties. In this paper, we propose a novel method to accurately segment natural images by texture and high order edge-detector. Firstly, we calculate local covariance matrix which is estimated from image gradient information within the local window, and use the eigenvalues of matrix to describe local texture feature of the image. Then, in order to suppress the effect of perplexing background, we introduce a high order edge-detector which can eliminate the background as much as possible while it can save the object boundary. Finally, the intensity term, texture term and edge-detector term are incorporated into the level set method to segment natural images. The proposed method has been tested on many natural images, and experimental results show the segmentation performance of the proposed method is better than prior similar state-of-the-art methods.