Object Detection Based on Shape Feature of Combined Regions

An object detection algorithm based on shape feature of combined regions is presented. Object candidates are first extracted by combining segmented regions, and then their feature vectors which contain angle chain and ratio of arc-chord distance to chord length are obtained. The parallel support vector machines are trained. The proposed algorithm requires low segmentation quality and can detect objects in blurred or low contrast images. The experiment on ETHZ Shape Classes demonstrates the effectiveness of the proposed algorithm.