A novel relative orientation feature for shape-based object recognition

Abstract

We propose a novel relative orientation feature (ROF) to represent the contour or skeleton of a two-dimensional object. With the aid of ROF, the shapes of two objects with fine structures can be compared. Matching with ROF is invariant with respect to translation, rotation and scaling transforms. Experimental results on hand gesture recognition demonstrate the effectiveness and efficiency of ROF with the identification rate of 98% and the average computational time less than 0.45ms/frame.