PROJECT TITLE :

Tag Refinement for User-Contributed Images via Graph Learning and Nonnegative Tensor Factorization

ABSTRACT:

Social image tagging systems mostly suffer from poor performance for Image Retrieval because of the noisy and incomplete correspondences between user-contributed pictures and their associated tags. During this letter, we have a tendency to aim to refine tag allocations within the social tagging data provided by these systems. In particular, we have a tendency to propose to harness the tagged and untagged information with a 2-stage strategy according to completely different varieties of information relations, i.e. item similarity outlined by prior information and item co-incidence learned from data statistics. To unravel the sparsity downside, we have a tendency to first introduce a new graph learning (GL) method for enriching the tagging knowledge in line with item similarities. Then, we tend to develop a method of nonnegative tensor factorization (NTF) for learning more coherent ternary relations among users, images and tags coupled by the manifold constraints learned from item co-occurrences. Experimental results with the tagging data from the NUS-WIDE dataset have been reported to validate the effectiveness of the proposed methodology.