In recent huge image databases, similarity-based retrieval technology, which enables users to retrieve similar images to the retrieval key in terms of image contents, becomes more and more essential. The database we will deal with in this paper is consisted of satellite images, especially cloud in those images, and the objective is to build a similarity-based retrieval system whose criteria are based on the similarities of both shape and spatial distribution of cloud regions. First the shape of a cloud region is represented using shape decomposition by affine-related polygons; then the decomposition result is structured into a hierarchical attributed relational graph. Under this formulation, graph matching cost is considered to be a natural similarity measure. The system, moreover, learns subjective evaluation by adjusting weighting factors associated with graph matching cost. All the modules are formulated based on optimization principle.

      @InProceedings{ kt:pru94,
      author = {Asanobu KITAMOTO and Mikio TAKAGI},
      title = {Similarity Retrieval of Satellite Cloud Imagery Based on Optimization Principle},
      booktitle = {Technical Report of Institute of Electronics, Information, and Communication Engineers (IEICE)},
      volume = {PRU94-49},
      pages = {15-22},
      year = 1994,
      month = 10,
      doi = {},
      note = { (in Japanese)},
      }