A Locality Sensitive Low-Rank Model for Image Tag Completion
Abstract of A Locality Sensitive Low-Rank Model for Image
A Locality Sensitive Low-Rank Model for Image Tag Completion,Many visual applications have benefited from the outburst of web images, yet the imprecise and incomplete tags arbitrarily provided by users, as the thorn of the rose, may hamper the performance of retrieval or indexing systems relying on such data. In this paper, our method draws inspiration from Multi-Task Learning (MTL) and formulates the local models by low-rank matrix factorization.
Such a model is able to promote information sharing between related tags as well as similar images.
However, it is not preferable to learn local models independently, since the output of data partition is typically far from satisfactory, even with the help of the pre-processing module.
As a result, the local models learned independently tend to overfit the data restricted to individual regions.
Therefore, to relieve the risk of overfitting as well as to promote robustness of the proposed LSLR method, a global consensus model is introduced to regularize the local models.
Conclusion
In this paper we propose a locality sensitive low-rank model for image tag completion.
The proposed method can capture complex correlations by approximating a nonlinear model with a collection of local linear models.
Our method achieves superior results on three datasets and outperforms pervious methods by a large margin.
