We have developed a practical scheme to take advantage of local typeface homogeneity to improve the accuracy of a character classifier. Given a polyfont classifier which is capable of recognizing any of 100 typefaces moderately well, our method allows it to specialize itself automatically to the single --- but otherwise unknown --- typeface it is reading. Essentially, the classifier retrains itself after examining some of the images, guided at first by the preset classification boundaries of the given classifier, and later by the behavior of the retrained classifier. Experimental trials on 6.4M pseudo-randomly distorted images show that the method improves on 95 of the 100 typefaces. It reduces the error rate by a factor of 2.5, averaged over 100 typefaces, when applied to an alphabet of 80 ASCII characters printed at ten point and digitized at 300 pixels/inch. This self-correcting method complements, and does not hinder, other methods for improving OCR accuracy, such as linguistic con...

The MetaPost system implements a picture-drawing language very much like Knuth’s META-FONT except that it outputs PostScript commands instead of run-length-encoded bitmaps. MetaPost is a powerful language for producing figures for documents to be printed on PostScript printers. It provides easy access to all the features of PostScript and it includes facilities for integrating text and graphics. This document serves as an introductory user’s manual. It does not require knowledge of METAFONT or access to The METAFONTbook, but both are beneficial. An appendix explains

The accuracy of today's document recognition algorithms falls abruptly when image quality degrades even slightly. In an effort to surmount this barrier, researchers have in recent years intensified their study of explicit, quantitative, parameterized models of the image defects that occur during printing and scanning. I review the recent literature and discuss the form these models might take. I give a preview of a large public--domain database of character images, labeled with ground--truth including all defect model parameters, the first of its kind. I describe the use of massive pseudo-randomly generated training sets for the construction of high-performance decision trees for preclassification. Also, I report preliminary results along a more theoretical line of attack: the estimation of the intrinsic error rate of precise--specified text recognition problems (this is joint work with Tin K. Ho). Finally, I list some open problems. 1 Introduction In recent years, some researchers i...

. The literature on models of document image degradation is reviewed, and open problems are listed. In response to the unpleasant fact that the accuracy of document recognition algorithms falls drastically when image quality degrades even slightly, researchers in the last decade have intensified their study of explicit, quantitative, parameterized models of image defects that occur during printing and scanning. Several models have been proposed, some motivated by the physics of image formation and others by the surface statistics of image distributions. A wide range of techniques for estimating parameters of these models has been explored. These models, in the form of pseudo-random generators of synthetic images, permit, for the first time, investigations into fundamental properties of concrete image recognition problems including the Bayes error of problems and the asymptotic accuracy and domain of competency of classifier technologies. The use of massive sets of syntheti...

user’s manual

Abstract. The research goal of highly versatile document analysis systems, capable of performing useful functions on the great majority of document images, seems to be receding, even in the face of decades of research. One family of nearly universally applicable capabilities includes document image content extraction tools able to locate regions containing handwriting, machine-print text, graphics, line-art, logos, photographs, noise, etc. To solve this problem in its full generality requires coping with a vast diversity of document and image types. The severity of the methodological problems is suggested by the lack of agreement within the R&D community on even what is meant by a representative set of samples in this context. Even when this is agreed, it is often not clear how sufficiently large sets for training and testing can be collected and ground truthed. Perhaps this can be alleviated by discovering a principled way to amplify sample sets using synthetic variations. We will then need classification methodologies capable of learning automatically from these huge sample sets in spite of their poorly parameterized—or unparameterizable—distributions. Perhaps fast expected-time approximate k-nearest neighbors classifiers are a good solution, even if they tend to require enormous data structures: hashed k-d trees seem promising. We discuss these issues and report recent progress towards their resolution.

Document image compression reduces the storage requirements for digitised books or documents by using characters as the fundamental unit of compression. Compression gains can be achieved by identifying regions that contain text, isolating unique characters, and storing them in a codebook. This thesis investigates several fundamental areas of the compression process. Algorithms for each area are tested on a corpus of images and the improvements tested for statistical significance. Methods for isolating characters from a bitmap are investigated along with techniques for determining reading order. We introduce the use of the docstrum to aid image compression and show that it improves upon previous methods. The Hough transform is shown to be an accurate method for determining page skew and gives robust results over a range of image resolutions. Compression is shown to improve when the skew of an image is determined automatically, and used to determine reading order. If images can be segm...

Abstract not found

First, a survey on optical scaling is carried out, both from the traditional point of view and from that of today’s digital typography. Then the special case of large characters, such as braces or integral signs, is considered. It is shown that such variable sized symbols should be computed at print time in order to approach the quality of metal typesetting. Finally, an implementation of such dynamic fonts, still in progress in the Grif editor, is described.

. We give a fast linear-time algorithm for finding a smooth polygonal approximation to a digitized contour such that the digitization of the polygonal contour matches the original input. The polygonal contour has the minimum possible number of inflections and obeys a localized best-fit property. Most of the vertices lie on a grid whose resolution is only twice that of the pixel grid, and the algorithm can be modified to force all vertices to obey this restriction. Key Words. smoothing, digitized image, polygonal outline, polygonization 1. Introduction. In graphics and computer typesetting, black and white images are commonly represented as arrays of pixels, and it is convenient to manipulate such images via the contours that describe black-white boundaries. For instance, if pixels are thought of as unit squares that tile the plane, the contours that describe the digitized image of a letter "R" might appear as in Figure 1a. Of course, the smoother contours shown in Figure 1b are a muc...