Image indexing and similarity retrieval based on A new Spatial Relation Model

Spatial relation model is important technique for image indexing and retrieved in image or multimedia databases. The 2-D strings and its variants are proposed to support the representation of spatial relationship. In this paper, a new spatial knowledge representation model named "Two Dimension Begin-End boundary string" (2D B epsilon -string) is proposed. The 2D B epsilon -string represents an icon by its MBR boundaries. By applying a number of "dummy objects", the 2D B epsilon -string can intuitively and naturally represent the pictorial spatial information without any special operator. In addition, an image similarity evaluation method based on the modified "Longest Common Subsequence" (LCS) algorithm is presented. By the proposed evaluation method, not only those images which all of the icons and their spatial relationships fully accord with the query image can be sifted out, but also for those images which partial of icons and/or spatial relationships are similar with the query image. It resolves the problems that query targets and/or spatial relationships are not certain. Our representation model and similarity evaluation also simply the retrieval progress of linear transformations, include rotation and reflection, of an image.