Automatic image annotation based on semi-supervised probabilistic CCA

Abstract

We propose a novel semi-supervised method for building a statistical model that represents the relationship between images and text labels (tags) based on a semi-supervised variant of CCA called Semi- PCCA, which extends the probabilistic CCA model to make use of the labelled and unlabelled images together to extract the low-dimensional latent space representing topics of images. Real-world image tagging experiments indicate that our proposed method improves the accuracy even when only a small number of labelled images are available.