The objective of this paper is to construct NOAA satellite image database that provides similarity retrieval measures. In particular, cloud patterns in satellite images are analyzed as natural objects. In this paper, a new model -- a hierarchical attributed relational graph (HARG) -- is proposed to represent spatial information, and another new model -- shape decomposition by affine equivalent polygons -- is proposed to represent shape information. Similarity is measured by the graph matching technique of HARGs. Finally, this similarity metric learns human subjectivity by adjusting weighting factors used in graph matching.

      @MastersThesis{ k:mastersthesis, 
      author = {Asanobu KITAMOTO},
      title = {Similarity Retrieval of Natural Images by Integrating Spatial and Shape Information}, 
      school = {東京大学工学系研究科電子工学専攻},
      year = {1994}, 
      month = {3}, 
      note = { (in English)}, 
      }