Recognition for Objects by Relationships Between Attributes

Authors

  • Hiroka Horiguchi Graduate School of Engineering, Soka University
  • Kazuo Ikeshiro Graduate School of Engineering, Soka University
  • Hiroki Imamura Graduate School of Engineering, Soka University

Keywords:

Face recognition, Robust recognition, Relationships between attributes, Occlusion, Irrelevant attributes.

Abstract

Object recognition methods based on attributes have been studied. Conventional methods recognize objects by the presence or absence of attributes. However, the conventional methods have two problems. Firstly, the conventional methods are not able to recognize a target object of which a part of attributes is occluded. Secondly, the conventional methods miss-recognize a target object, which has irrelevant attributes. Therefore, to solve these two problems, we propose the object recognition by relationships between attributes. In this paper, we focus on the face as the recognition object. The proposed method uses relationships as constraints for object recognition using attributes. The proposed method applies two major type constraints. The first constraint is a local constraint, which is applied to a part of attributes. To achieve robust face recognition against occlusion scenes, the proposed method uses the local constraint. And then, the second constraint is a global constraint, which is applied to all attributes. To achieve robust face recognition against irrelevant attributes, the proposed method uses the global constraint. In this paper, to evaluate the effectiveness of the proposed method, we compared the proposed method with a conventional method. We experimented in normal face, occlusion and irrelevant attributes. We used 2580 images of a face which are changed in scale and rotation. Experimental results showed that the recognition ratio of the proposed method is equal to or more than that of the conventional method in normal face, occlusion, and irrelevant attributes.

Downloads

Published

2016-12-26

Issue

Section

Articles