A new paradigm in image indexing and retrieval using composite bitplane signatures

Abstract

A scheme based on bitplane signatures for fast image searching is presented. It makes use of composite bitplanes obtained from overlaying bitplane signatures to form composite bitplane signatures. For large image databases, a hierarchical structure of composite bitplane signatures can be used to provide ranked images without the need to compare and traverse all the images in the database, as many existing retrieval methods do. To search a particular signature embedded in a composite bitplane signature, we rank all the signatures, not by the number of matched bits but by the least difference of 1-bits from the query signature. Experiments are performed using 1,000 images to observe the behavior of composite bitplane signatures under wavelets, and we have also determined the most suitable number of images for insertion into a composite bitplane signature for its efficient operation.