Multimodal Automatic Image Annotation Method using Association Rules Mining and Clustering

Effective and fast retrieval of images from image datasets is not an easy task, especially with the continuous and fast growth of digital images added everyday by used to the web. Automatic image annotation is an approach that has been proposed to facilitate the retrieval of images semantically related to a query image. A multimodal image annotation method is proposed in this paper. The goal is to benefit from the visual features extracted from images and their associated user tags. The proposed method relies on clustering to regroup the text and visual features into clusters and on association rules mining to generate the rules that associate text clusters to visual clusters. In the experimental evaluation, two datasets of the photo annotation tasks are considered; ImageCLEF 2011 and ImageCLEF 2012. Results achieved by the proposed method are better than all the multimodal methods of participants in ImageCLEF 2011 photo annotation task and state-of-the-art methods. Moreover, the MiAP of the proposed method is better than the MiAP of 7 participants out of 11 when using ImageCLEF 2012 in the evaluation.