Automatic image annotation via label transfer in the semantic space

Highlights：

• Automatic image annotation framework based on label propagation in the semantic space.

• We use Kernel Canonical Correlation Analysis to build a latent semantic space where correlation of visual and textual features are well preserved.

• The approach is robust and work either when the training set is annotated by experts, as well as when it is noisy such as in the case of tags in social media.

• We report extensive results on four popular datasets.

• Our results show that our KCCA-based framework can be applied to several state-of-the-art label transfer methods to obtain significant improvements.