A Region-of-Interest Method for Texturally-Rich Document Image Coding X-W. Yin, A. C. Downton, M. Fleury, and J. He Department of Electronic Systems Engineering, University of Essex, United Kingdom Tel. +44 (0)1206 872817 Fax. +44 (0)1206 872900 fleum@essex.ac.uk Abstract – Region-of-Interest (ROI) techniques are often utilized to improve coding for detailed regions in natural still-image coding standards such as JPEG2000 [1], but no specific method is stated for determining the ROI map. In this paper, an ROI-based method, in which rectangular regions are extracted using document image analysis (DIA), is proposed specifically for document image coding. These rectangular regions can be efficiently coded using wavelets, and DIA may also be used to distinguish between important and unwanted foreground regions, allowing further coding gains (as illustrated in one of the example documents in the paper). Compared to multi-layer methods currently used for document image coding [2], the method is simpler and scalable, while improving visual quality and the Peak-Signal-to-Noise Ratio (PSNR). EDICS category – 2.IMMD--Image and Multidimensional Signal Processing I. INTRODUCTION Historically, the most common format for document images has been binary for reasons of efficient storage, leading to the development of binary document image coding standards such as JBIG1 and JBIG2 [3], based upon run-length coding techniques. However, as demand for higher image quality has grown and the range of digitized documents increased, gray-scale and color document image representations have become common, although these increase storage space and/or transmission time. Hence, it is now essential to design document image coding algorithms that can compactly represent texturally-rich document images, which are increasingly being made accessible through online document archives [4] (see [5] for an example used in this paper). The main contribution of this paper is to point out the advantages of using rectangular ROI-based compression on document images, in terms of algorithmic simplicity and accuracy of representation. By way of illustration, a case study showing extraction of ROIs for differing types of textual regions is demonstrated, using a low-complexity wavelet codec [6] adapted for ROI extraction. The paper makes no special claims for the DIA techniques applied, which would typically be adapted to the type of document archive being coded. However, the paper does demonstrate that the rectangular-ROI technique, combined with bit-plane shifted wavelet coding, is very competitive in both objective and subjective visual quality with current multi-layer document coding techniques [2]; it is also considerably more computationally efficient. 1