A Modified Sobel Edge Detection Using Dempster-Shafer Theory

A modified Sobel edge detection is proposed in this paper. Dempster-Shafer theory, also known as the theory of belief function, is applied to improve the drawbacks of the conventional Sobel operator, for instance, the thick edge and sensitive to noise. The reason is that by selecting the mass function, Dempster-Shafer theory can distinguish the edge pixels from the uncertain edge pixels correctly, which can suppress the noise and make the edge thinner. Firstly, the Sobel operator is employed to obtain the gradient magnitude G_x and G_y, which are regarded as the independent sources; and the mass function is selected by the gradient values and the overlapping surface of the constructed triangles; then, the orthogonal sum is calculated; finally, the mass function of the edge probability is taken as the edge image. From the experiment, the edge is thin and noise-free, so the result could be accepted.