3D Model-Based Sematic Labeling of 2D Objects

This paper tackles the issue of still image object categorization. The objective is to infer the semantics of 2D objects present in natural images. The principle of the proposed approach consists of exploiting categorized 3D synthetic models in order to identify unknown 2D objects, based on 2D/3D matching techniques. Notably, we use 2D/3D shape indexing methods, where 3D models are described through a set of 2D views. Experimental results carried out on both MPEG-7 and Princeton 3D mesh test sets show recognition rates of up to 89%.