A level set model by regularizing local fitting energy and penalty energy term for image segmentation

摘要

A novel level set model is proposed by regularizing local fitting energy to segment the intensity inhomo-geneous images. In proposed model, local image fitting energy information is incorporated with penalty energy function. To segment intensity inhomogeneous images, a circular window filter is used to identify homogeneous regions. In addition, local information of homogeneous regions is constructed in form of local fitting function. A new double well potential function is formulated to regularize the contour curve and is used a penalty energy term which is added with local fitting functional in order to formulate multi-scale energy function. The experimental results show the robustness of proposed method compared to other level set models and deep learning based level set model. The comparative study of Jaccard similarity index (JSI) values and segmentation accuracy also validate the preciseness of the proposed model. Further, the proposed model yields better segmentation results compared to the other state-of-the-art models in terms of higher precision and recall values and lesser computational time. In addition, the proposed model is computationally efficient and robust to noise as well as contour initialization.