Edge Detection Based on Adaptive Oriented Double Opponent Neurons

摘要

A new method of edge detection based on adaptive oriented double opponent neurons is presented in this paper, considering that the homocentric opponent receptive field is lack of directionality, and the anisotropic of receptive field in ODOG model will be badly restrained during weighting process. To get relatively complete image edges, the edge directional operators are introduced to choose Difference of Gaussians (DOG) model or the orientations of Oriented Difference of Gaussians (ODOG) model automatically. Compared with DOG and ODOG methods, the methods detect weak edges effectively with better edge connectivity and edge confidence.