Detecting Edges with Bi-Directional Tracing Across Multiple Scales

For our human vision system, natural images are understood via a progressive and hierarchical process; that is, a multi-scale analysis. Motivated by this fact, a multi-scale algorithm for edge detection is proposed in this paper. The novelty of the proposed algorithm lies in that a bi-directional tracing strategy is used, which first conducts a forward tracing (FT) operation in coarse-to-fine direction and then a backward confirm (BC) operation in fine-to-coarse direction across multiple scales. With the FT operation, the location accuracy of detected edges is improved; with the BC operation, spurious edges are removed. The edge map generated with this bi-directional tracing strategy can therefore have a better quality than that generated by the existing multi-scale edge detectors. Extensive simulation results have shown that our new algorithm is superior to a number of state-of-the-art edge detection algorithms in both subjective visual assessment and objective measurement.