Image Indexing and Similarity Retrieval Based on A New Spatial Relation Model

Spatial relation model is important technique fo r iniage indexing arid retrieval in iniage or niultiniedia databases. 2 0 strings and its variants are proposed to s~rpport the representation of spatial relationship. I n this paper, a new spatial knowledge representarion model rianied “Tvi.0 Diniension Begin-End boundan string”(2D BE-string) is proposed. The 2 0 BE-string represents on icon by its MBR boundaries. By applying a nuniber of “duniniy objects”, the 2 0 BE-string can intuitively atid tiat~rrally represent the pictorial spatial iriforniatiori withoirt any spatial operator. In addition, an image siniilarity evaluation method based on the modified “Longest Common Subseq~ierice ” (LCS) algorithm is presented. By the proposed evaliiation method, not only those iniages \c.hicli all of the icons and their spatial relatiorishipsfitll~ accord with the q u e n image can be sfted out, birr also fo r those iniages which partial of icons widor spatial relationships are similar with the q u e n iniage. I t resolves the problenis hat the qrren targets andor spatial relationships are not certain. Our representation model arid siniilarih evaluation also simply the retrieval progress of linear transforniations, include rotation arid reflection, of on image. ’ Keyword: iniage retrieval, iniage database, spatial knowledge, spatial reasoning, similarity retrieval, 2 0 Strings, LCS algorithm, 2 0 Bestring