Multi-view semi-supervised learning for image classification.pdfVIP

Neurocomputing 208 (2016) 136–142 Contents lists available at ScienceDirect Neurocomputing journal homepage: /locate/neucom Multi-view semi-supervised learning for image classi?cation Songhao Zhu, Xian Sun, Dongliang Jin School of Automation, Nanjing University of Posts and Telecommunications, Nanjing 210046, China article info Article history: Received 17 September 2015 Received in revised form 10 December 2015 Accepted 8 February 2016 Available online 31 May 2016 Keywords: Multi-view semi-supervised learning Maximum entropy Multi-view fusion Image classi?cation abstract With the massive growth of digital image data uploaded to the Internet, classifying each image into appropriate semantic category with respect to its image content for image index and image retrieval has become an increasingly dif?cult and laborious task. To deal with this issue, we propose a novel multiview semi-supervised learning framework which leverages the information contained in pseudo-labeled images to improve the prediction performance of image classi?cation using multiple views of an image. In the training process, labeled images are ?rst adopted to train view-speci?c classi?ers independently using uncorrelated and suf?cient views, and each view-speci?c classi?er is then iteratively re-trained with respect to a measure of con?dence using initial labeled samples and additional pseudo-labeled samples. In the classi?cation process, the maximum entropy principle is utilized to assign appropriate category labels to unlabeled images via optimally trained view-speci?c classi?ers. Experimental results on a general-purpose image database demonstrate the effectiveness and ef?ciency of the proposed multi-view semi-supervised image classi?cation scheme. 2016 Elsevier B.V. All rights reserved. 1. Introduction With the explosive growth of digital image collections on the internet, the demand of developing content-based analysis technologies to effectively organize, manage and utilize such huge amount