Learning robust multi-label hashing for efficient image retrieval
Supervised hashing generally achieves superior performance over unsupervised or semi-supervised approaches by leveraging semantic labels. However, most existing supervised hashing techniques only deal with image samples with single label. Few of them properly address the practical problem concerning...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | text |
| Language: | English |
| Published: |
Institutional Knowledge at Singapore Management University
2016
|
| Subjects: | |
| Online Access: | https://ink.library.smu.edu.sg/sis_research/3593 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Singapore Management University |
| Language: | English |
| Summary: | Supervised hashing generally achieves superior performance over unsupervised or semi-supervised approaches by leveraging semantic labels. However, most existing supervised hashing techniques only deal with image samples with single label. Few of them properly address the practical problem concerning images with multiple labels, which is very common in real applications. In this paper, we seek to address the limitations of the existing schemes by proposing a novel approach, dubbed as Robust Multi-Label Hashing (RMLH). A label hypergraph is constructed to effectively capture high-order semantic correlations of images. And they are preserved into hashing codes with hypergraph consistency and direct label-hashing correlation. Besides, we impose a nuclear norm regularization on correlation matrix to maintain label correlations and robustly accommodate missing labels. Furthermore, an efficient algorithm based on Alternate Direction Method of Multipliers (ADMM) is developed to calculate the optimal hashing codes. Experiments demonstrate that RMLH can outperform state-of-the-art schemes and enjoy much better robustness against missing labels. |
|---|