Object categorization in context based on probabilistic learning of classification tree with boosted features and co-occurrence structure

Abstract

This paper proposes a probabilistic method of object categorization in context through learning a classification tree with boosted features and co-occurrence structure. In this method, object classes are obtained for each scene category, a classification tree with boosted features is generated for all the object classes and co-occurrence is analyzed among object categories in scenes. In recognition, object categories in a scene are simultaneously determined based on a classification tree search using composite boosted features under co-occurrence constraint and a foreground object is inferred based on object category composition of scene categories. Through experiments using images of plural categories in an image data set, it is shown that object categorization performance is improved by using boosted features and co-occurrence structure, especially by using both of them. © 2013 Springer-Verlag.