Inter-media Pseudo-relevance Feedback Application to ImageCLEF 2006 Photo Retrieval

摘要

This paper shows how to use a text retrieval and an image retrieval engine in a cooperative way. The proposed Inter-Media Pseudo-Relevance Feedback approach shows how the image modality can be used for expanding the text query and significantly improve the performance (MAP) of a multimedia information retrieval system. The document collection used for experiments is provided by the International Association for Pattern Recognition (IAPR) [1]. The proposed approach is completely automatic and has been used to participate to the ImageCLEF 2006 Photo Retrieval task.