Multimodal subspace learning on Flickr images

Abstract

Extensive computational power and the substantial research in the field of image processing and feature extraction automatically generate a need of knowledge discovery from multiple modalities of images. Large number of Flickr images are available and various knowledge discovery research techniques have been applied on this dataset for creating intelligent decisions. Given different views of objects, finding a unique latent semantically fused view is one of the challenging tasks in the field of Machine Learning and Pattern Recognition. In this paper, we propose to use a matrix factorization method to learn a semantic subspace using multiple feature sets for Flickr dataset. We applied the matrix factorization technique on NUS-WIDE Object dataset, and achieved better accuracy than the previous approaches on the same dataset. Moreover, the training time for our technique is significantly reduced. Results show the efficacy of the proposed framework. © 2015 IEEE.