Multilevel Indexing Structure for Object Based Image Retrieval

Object based technique is a good way to grasp the user´s query intentions felicitously and at the same time to make retrieval procedure flexible with user to avoid pseudo retrieval in content-based image retrieval (CBIR). But the main problem for such technique is its high computation complexity for searching objects among image dataset. Aiming at the problem mentioned above, a multilevel indexing structure (MIS) for object-based image retrieval is proposed in this paper. A three-level tree structure in the image domain is first constructed to reduce the object searching complexity. In order to further fasten the speed of scanning objects, i.e., to eliminate those unrelated images in the dataset rapidly, the clustering in the dataset domain is also conducted. The final experimental results show that the proposed MIS for image retrieval is much more efficient