Connected Edge Ends for Object Recognition

In this paper we exploit edges in images to build features for object category recognition. We build a line based image approximation to identify the dominant edge structures and use SIFT descriptors to capture the object shape. Combinations of co-occurring features improve the recognition performance but suffer from complexity due to combinatorial problems. To represent more complex structures we combine descriptors connected by lines fitted to dominant edges. In this way we incorporate weak topological constraints into the representation which improve its discriminative