Indexing of Document Images Based on Triangular Spatial Relationships

Abstract

In this paper, a new scheme of indexing document images based on B-tree by preserving triangular spatial relationships (TSR) among the components of a document image is proposed. A new technique for labeling of components in document images is also proposed. A procedure for classifying retrieved images and a method of ranking the retrieved document images are also proposed. The retrieval results of the proposed TSR based indexing scheme is compared with nine-directional codes based (NDC) indexing scheme and also with the retrieval results of human experts. The experiments are conducted on the MediaTeam document image database that provides diverse collection of document images