Boundary points handling for image edge detection based on Neutrosophic set

In this work, a new approach based on Neutrosophic set (NS) is proposed for image edge detection. At first, a new definition for boundary points in NS domain is presented. Then, a new cost function is designed to separate boundary points from other points. Finally, each data point is assigned to main, boundary and outlier clusters with 3 membership degrees. After these assignments, points with higher membership degrees to boundary cluster are considered as edge pixels. The proposed method is evaluated in two types of image datasets including artificial and real images. Results demonstrated that the proposed method detects edges more efficiently and accurately in comparison to state of the art edge detection algorithms.