A descriptive retrieval engine for image databases

Electronic images are rapidly replacing conventional paper documents due to their handling advantages. As a result, there is a need for retrieval techniques where the level of man-machine interaction is closer to that enjoyed amongst the humans. This paper describes the work carried out to date in design and development of a descriptive retrieval engine. In such a system users may retrieve documents from an unorganised image database using “high level” descriptions, such as: “...I´m looking for a tax form. It was mainly blue with a customs and excise logo on top right hand corner. I think it was a 95-96...” The automatic information extraction part of the system has been successfully tested on many widely varying documents segmenting them into pure text, pure graphic and mixed entities