Imitating the Biological Vision Mechanism for Orientation Estimation

Abstract

Local orientation is a kind of features of multi-dimensional signals. Orientation estimation play an important role in many image processing and computer vision tasks. This paper puts forward a revised orientation estimation algorithm called orientation bar chart for estimating local orientation imitating the biological vision mechanism described by the Geatalt law. The method calculates the vector addition of all neighboring voting for an object image pixel to get the orientation's bar chart and search out the peak value of the bar chart as its orientation angle. These orientation angles are used for later tensor voting process as orientation infor-mation. Two group of experiments are performed on typical contours to compare the contour completion effect and the accuracy. Experimental results on edge graph of city road show that the proposed algorithm works well for improving contour intaction, producing less clutter pixels and being robust to noises.