Abstract
Automatic image annotation is an important technique which has been widely applied in many fields such as social network image analysis and retrieval, face recognition and so on. Multi-view image annotation aims to utilize multi-view complementary information to achieve more effective annotation results. However, the existing multi-view image annotation methods cannot well handle the complex and diversified multi-view feature, and the label correlation is also ignored. In this paper, we propose an image annotation method by integrating deep multi-view latent space learning and label correlation guided image annotation into a unified framework, which is termed as Label Correlation guided Deep Multi-view image annotation (LCDM) method. LCDM first learns a consistent multi-view representation via deep matrix factorization, which well captures multi-view complementary information. Then, label correlation is exploited to improve the discriminating power of the classifiers. We propose a unified objective function so that multi-view data representation and classifiers can be jointly learned. Extensive experimental results on various image datasets demonstrate the effectiveness of our method.
Highlights
A large number of image data are uploaded and disseminated on social network platforms every day
How to bridge the gap between low-level visual features and high-level semantics is the key of image annotation, and various image annotation methods have been developed in recent years
We propose a Label Correlation guided Deep Multi-view image annotation method (LCDM) to predict image labels based on multi-view features
Summary
A large number of image data are uploaded and disseminated on social network platforms every day. The most common way of image annotation is to adopt multi-label learning method to predict image labels [6], [7]. It jointly factorizes multiple matrices so that multi-view feature and image labels can be associated in a consistent latent space. By maximizing the correlations between multiview feature space, label space and latent space, it can effectively predict image labels with many classes. We propose a Label Correlation guided Deep Multi-view image annotation method (LCDM) to predict image labels based on multi-view features. Our method first learns deep multi-view latent space via deep multiview matrix factorization model to represent multi-view data, which effectively encodes the multi-view complementary information into a unified latent space. Based on the learned deep multi-view latent space, label classifiers are learned to predict image labels. Our method further improves the discriminating power of the classifiers
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
