Document Image Coding for Processing and Retrieval

Abstract

Document images belong to a unique class of images where the information is embedded in the language represented by a series of symbols on the page rather than in the visual objects themselves. Since these symbols tend to appear repeatedly, a domain-specific image coding strategy can be designed to facilitate enhanced compression and retrieval. In this paper we describe a coding methodology that not only exploits component-level redundancy to reduce code length but also supports efficient data access. The approach identifies and organizes symbol patterns which appear repeatedly. Similar components are represented by a single prototype stored in a library and the location of each component instance is coded along with the residual between it and its prototype. A representation is built which provides a natural information index allowing access to individual components. Compression results are competitive and compressed-domain access is superior to competing methods. Applications to network-related problems have been considered, and show promising results.