Adaptive hierarchical contexts for object recognition with conditional mixture of trees

Abstract

Robust category-level object recognition is currently a major goal for the computervision community. Intra-class and pose variations, as well as, background clutter andpartial occlusions are some of the main difficulties to achieve this goal. Contextual in-formation, in the form of object co-occurrences and spatial constraints, has been suc-cessfully applied to improve object recognition performance, however, previous workconsiders only fixed contextual relations that do not depend of the type of scene underinspection. In this work, we present a method that learns adaptive conditional relation-ships that depend on the type of scene being analyzed. In particular, we propose a modelbased on a conditional mixture of trees that is able to capture contextual relationshipsamong objects using global information about a scene. Our experiments show that theadaptive specialization of contextual relationships improves object recognition accuracyoutperforming previous state-of-the-art approaches.