Structure-preserving document image compression

Maintaining a document in image form is often preferable in order to avoid the high cost of manual conversion or the introduction of large numbers of errors by automatic OCR and/or graphics interpretation. The large volume of data in the image can be greatly reduced by using compression techniques. Text-intensive document images typically have a great deal of redundancy in the bitmap representations of symbols, and we make use of that redundancy for compression by clustering components, representing each cluster by a template and encoding the error. Our method is novel in modeling the error associated with each cluster and in preserving structure, an important component for readability and processing. Document image compression for storage or transmission is an important research topic, in part because of the relatively large amounts of data that documents contain. Text-intensive document images typically have signiicant amounts of symbol-level redundancy as characters of the same size and font repeat in the text. Our research centers around exploiting the redundancy to provide a compressed representation which preserves document-level features necessary for later processing and analysis 5]. Many image compression schemes are widely available. For binary document images, CCITT Group 3 and 4 are among the industry standards 8]. These standards are lossless compression schemes which were proposed for fax and modem use. They use variations of run-length encoding for optimized binary image transmission. For grayscale and color images, the JPEG standard 6] is a widely accepted lossy compression scheme. This method works well with textured images but does not exploit the structural redundancies within a document. In general, methods of image (a) (b) Figure 1: a) Symbolic image, b) Error image coding based on image models (e.g., 2]) do not exploit the redundancies that exist in documents and may render a document unreadable. Since many document processing algorithms degrade at resolutions below 300 dpi, simply reducing the resolution to save space is not necessarily an option. An important consideration is the preservation of structure so that symbols remain recognizable. In this paper we use a structural approach to compress textual images. The general approach, suggested in part by Ascher and Nagy 1] and later enhanced by Witten 7], rst clusters text-like components and represents each cluster by a single template (presumably of the underlying symbol); the residual error is then encoded (Figure 1). Our contribution is the use of a probabilistic model of the errors. In …