An Adaptive Canny Detector with New Differential Operator

A new Canny detector is proposed, which adopts both new differential operator and adaptive selection of double thresholds. One contribution of this paper is the new differential operator which evolves from quadratic Taylor expansion. It has the advantages of first-order differential operator and second-order differential operator. The other contribution is that double thresholds in Canny detector are selected automatically by using Otsu´s method based on gradient magnitude histogram. Experiment results show that the new Canny detector could get more weak edges and even locate these edges more accurately than the original one. Moreover, it could find optimal edges adaptively instead of manually. This new Canny detector has remarkable performance when it is applied to images with low contrast, or those images which pursue accurate location of edges.