Regularized extreme learning machine for multi-view semi-supervised action recognition.pdfVIP

Neurocomputing 145 (2014) 250–262 Contents lists available at ScienceDirect Neurocomputing journal homepage: /locate/neucom Regularized extreme learning machine for multi-view semi-supervised action recognition Alexandros Iosi?dis n, Anastasios Tefas, Ioannis Pitas Department of Informatics, Aristotle University of Thessaloniki, Box 451, 54124 Thessaloniki, Greece article info Article history: Received 14 May 2013 Received in revised form 19 December 2013 Accepted 12 May 2014 Communicated by L. Shao Available online 2 June 2014 Keywords: Extreme learning machine Semi-supervised learning Multi-view learning abstract In this paper, three novel classi?cation algorithms aiming at (semi-)supervised action classi?cation are proposed. Inspired by the effectiveness of discriminant subspace learning techniques and the fast and ef?cient Extreme Learning Machine (ELM) algorithm for Single-hidden Layer Feedforward Neural networks training, the ELM algorithm is extended by incorporating discrimination criteria in its optimization process, in order to enhance its classi?cation performance. The proposed Discriminant ELM algorithm is extended, by incorporating proper regularization in its optimization process, in order to exploit information appearing in both labeled and unlabeled action instances. An iterative optimization scheme is proposed in order to address multi-view action classi?cation. The proposed classi?cation algorithms are evaluated on three publicly available action recognition databases providing state-of-theart performance in all the cases. 2014 Elsevier B.V. All rights reserved. 1. Introduction Human action recognition from videos is receiving increasing attention, due to its importance in a wide range of applications, like intelligent visual surveillance, human–computer interaction and content-based video annotation/retrieval, to name a few. However, it is a challenging problem, because of the complexity of human actions. The dynamic human body motion patterns ca