A novel edge detection approach using a fusion model

Edge detection is a long standing but still challenging problem. Although there are many effective edge detectors, none of them can obtain ideal edges in every situation. To make the results robust for any image, we propose a new edge detection algorithm based on a two-level fusion model that combines several typical edge detectors together with new proposed edge estimation strategies. At the first level, we select three typical but diverse edge detectors. The edge score is calculated for every pixel in the image based on a consensus measurement by counting positive voting number of approaches. Then results are combined at the second level using the Hadamard product with two additional edge estimations proposed in the paper, based on edge spatial characteristics, where one is binary matrix of the most probable edge distribution and the other is a score matrix based on calculating differences between maxima and minima neighboring intensity change at each point. Comprehensive experiments are conducted on two image databases, and three evaluation methods are employed to measure the performance, viz. F1-measure, ROC and PFOM. Experiments results show that our proposed method outperforms the three standard baseline edge detectors and shows better results than a state-of-the-art method.