Abstract: Contour extraction is a classical problem of computer vision.Based on the visual perception theory this paper presents an algorithm of multi-feature orientation preference.This algorithm is composed of three layers.Firstly,we separated luminance and color information.Then those information were filtered by the model of difference of Gaussians to simulate the concentric circle antagonism of ganglion cells.At last we simulated the visual processing mechanism of primary visual cortex.The significant edge information was obtained by directional filtering and then filtered by inherent variation,and finally integrated into contours.Experimental results show that this algorithm is fast in calculating speed and can extract complete contours with lower noise.