Tag propagation by using multi-view NMF consistent matrix for image annotation

Abstract

An image tag propagation method could assign tags to a new query image from labeled training samples by paired comparison of visual distances. One of its challenges is the inconsistency between visual similarities and tag similarities. In this paper, to benefit from the structure information commonly described by visual features and image tags, we propose a novel propagation based on multi-view Negative-Matrix-Factorization(NMF) clustering and sparse tag sensing. A ranked consistent matrix is created from multiview NMF observations to estimate the clustering structures, while sparse tags of a query image are approximately reconstructed with consistent matrix sensing. Compared to the best performing 2PKNN algorithm, our proposed method gains 3.2%, 0.8% in term of average recall and F1-score.