Hierarchical content description and object formation by learning

Efficient management of digital multimedia databases requires the storage, searching, browsing and retrieval of large amounts of visual information effectively. This paper proposes a hierarchical representation scheme and a hierarchical content matching technique that provide a learning mechanism for effective and efficient retrieval from user profiles and search patterns. The proposed representation, together with a low-level image analysis procedure, also enables the automatic analysis of images into meaningful objects for object-based content description. Experimental results are presented to show the effectiveness of the proposed representation scheme, the hierarchical matching technique and the learning concept.