Significant edge finding in view of human perception

a novel approach is presented to finding significant edges from noisy images, which is characterized by imitating two abilities of human in qualifying significant edges, namely edge extraction from heterogeneous or homogeneous objects, and to weight on edges with similar directions tending to align along a trajectory. Gradient directions are evaluated on selected pixels via entropy weighting, followed by employing a variable mask to scan the weighting results to identify alignments. Bayesian decision making scheme is used to exploit fine and coarse edge features. Simulation results are provided to show noise resistance and the capability of imitating human visual perception.