A Fast, Background-Independent Retrieval Strategy for Color Image Databases
Abstract
We describe an interactive, multi-phase color-based image retrieval system which is capable of identifying query objects specified by the user in an image in the presence of significant, interfering backgrounds. The system uses a split and merge histogram peak detection technique, efficient indexing and a color neighborhood graph-based refinement phase for removing false matches. The method is fast and has low storage overhead. Good retrieval results are obtained with multi-colored query objects even when they occur in arbitrary sizes, rotations and locations in the database images. The overall scheme is flexible so that one or more phases could be used for retrieval depending on the precision desired. The experimental results on a database of advertisement images highlights the capabilities of the system.
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 01, 1996
- Accession Number
- ADA478119
Entities
People
- Bruce A. Draper
- E. M. Riseman
- Mini Das
- R. Manmatha
- W. J. Lim
Organizations
- University of Massachusetts Amherst