2 System Design

An outline for the design of the IM³ system is presented in Figure 1. Specially modified digital photocopiers were developed that automatically capture an image of every copied document. Aside from a user identifying himself by pressing a button on a touchscreen, this process is completely transparent. The captured images are transferred to the document server where they are permanently stored and indexed for later retrieval.

Print jobs are automatically captured by software running on a Unix print server. A copy of every printed document is transferred to the document server as it is sent to a printer. This is done by a filter in the spooling system that is applicable to jobs printed on PC's, Apple computers, and Unix workstations. In this way the capture of printed documents is completely transparent to the user and is independent of any application software. Every document sent to printers serviced by the Unix server is saved.

An indexing process is applied to every saved document. The images from the photocopiers are OCR'd. Text is extracted from the postscript files for printed documents. This is used to choose keywords for each document and build data structures for full text retrieval. Thumbnail images are also calculated at several resolutions (4 dpi, 8dpi, and 72 dpi) for use in various browsable interfaces.