This paper proposes a ``hierarchical model,'' used as a framework for constructing similarity-based image retrieval systems. This model represents ``image content'' as a set of layers divided from a viewpoint of meaning. The proposed hierarchical architecture grasps lucid perspective structure of similarity-based image retrieval systems, and it becomes possible to apply the system, without changing the framework, to a wide range of problem domains by virtue of the design to independently "plug in" appropriate methods to each layer. Similarity-based image retrieval systems based on the hierarchical model are constructed for different types of imagery, namely two types of satellite imagery and landscape imagery, and the result shows that the proposed model works effective for multiple types of imagery.

      @InProceedings{ kt:prmu97,
      author = {Asanobu KITAMOTO and Mikio TAKAGI},
      title = {Hierarchical Model as a Framework for Constructing Similarity-Based Image Retrieval Systems},
      booktitle = {Technical Report of Institute of Electronics, Information, and Communication Engineers (IEICE)},
      volume = {PRMU97-58},
      pages = {25-32},
      year = 1997,
      month = 7,
      doi = {},
      note = { (in Japanese)},
      }