| |
| A novel method for image tag completion is proposed in this paper, which draws inspiration from multi-task learning and aims to promote information sharing between similar samples and related tags via low rank matrix factorization. Specifically, the initial tag matrix is decomposed into a basis matrix and a sparse coefficient matrix, and proper regularization is introduced to exploit various side information, by ensuring the preservation of local geometry structures in both tag space and feature space. Furthermore, the scheme of elastic net is utilized in our sparse construction, to gain improved robustness. Experiments conducted on two datasets with different features demonstrate the effectiveness and efficiency of the proposed method. |
|
| Keywords:Artificial intelligence, tag completion, multi-task learning, sparse coding. |
|
