Description and recognition of object contours using arc length and tangent orientation

We propose an object description and recognition approach based on the relationship between the arc length and tangent orientation angle of object contours in this paper. According to the approach, B-spline curves are firstly used to approximate contours to get the smooth variance of the arc length and tangent orientation angle, and then B-spline curves are divided into sets of simple patches, which compose the feature chains of contours. We also propose two description functions of contours, and discuss their invariance to translation, rotation and scaling transformations. In matching, because the features to be used are only several different patches of contours not the whole ones, our approach can be used to recognize not only objects with different scaling and poses, but also objects be partly occluded or absent. We conducted experiments with a set of airplane images and got a much high recognition rate.