PROJECT TITLE:

A Locality Sensitive Low-Rank Model for Image Tag Completion - 2016

ABSTRACT:

Many visual applications have benefited from the outburst of internet images, however the imprecise and incomplete tags arbitrarily provided by users, because the thorn of the rose, may hamper the performance of retrieval or indexing systems looking forward to such information. In this paper, we propose a completely unique locality sensitive low-rank model for image tag completion, that approximates the global nonlinear model with a assortment of native linear models. To effectively infuse the idea of locality sensitivity, a straightforward and effective pre-processing module is intended to learn appropriate illustration for data partition, and a international consensus regularizer is introduced to mitigate the chance of overfitting. Meanwhile, low-rank matrix factorization is used as native models, where the local geometry structures are preserved for the low-dimensional representation of each tags and samples. Extensive empirical evaluations conducted on three datasets demonstrate the effectiveness and potency of the proposed methodology, where our method outperforms pervious ones by a massive margin.