Automatic Image Annotation Using Modified Multi-label Dictionary Learning

Automatic image annotation has attracted lots of research interest, and effective method for image annotation. Find effectively the correlation among labels and images is a critical task for multi-label learning. Most of the existing multi-label learning methods exploit the label correlation only in the output label space, leaving the connection between label and features of images untouched. In image annotation, a semi supervised learning which incorporates a large amount of unlabeled data along with a small amount of labelled data, is regarded as an effective tool to reduce the burden of manual annotation. But some unlabeled data in semi-supervised models contain distance that negatively affects the training stage. Outliers in the method can be over-fitting problem especially when a small amount of training data is used. In this paper, proposing an automatic image annotation method called modified MLDL with hierarchical sparse coding for solving these problems. This method prevents the over-fitting associated with the semi-supervised based approach by using sparse representation to maximizing the correlation between the data. Apply a Tree Conditional Random Field to construct the Hierarchical structure of an image. The result will be multi-label set prediction of a query image and semantic retrieval of images. Experiment results using LabelMe datasets and Caltech datasets confirms the effectiveness of this method.