Most offline handwriting recognition approaches proceed by segmenting words into smaller pieces (usually characters) which are recognized separately. The recognition result of a word is then the composition of the individually recognized parts. Inspired by results in cognitive psychology, researchers have begun to focus on holistic word recognition approaches. Here we present a holistic word recognition approach for single-author historical documents, which is motivated by the fact that for severely degraded documents a segmentation of words into characters will produce very poor results. The quality of the original documents does not allow us to recognize them with high accuracy - our goal here is to produce transcriptions that will allow successful retrieval of images, which has been shown to be feasible even in such noisy environments. We believe that this is the first systematic approach to recognizing words in historical manuscripts with extensive experiments. Our experiments show a recognition accuracy of 65%, which exceeds performance of other systems that operate on non-degraded input images (non historical documents) .

Searching and indexing historical handwritten collections is a very challenging problem. We describe an approach called word spotting which involves grouping word images into clusters of similar words by using image matching to find similarity. By annotating “interesting ” clusters, an index that links words to the locations where they occur can be built automatically. Image similarities computed using a number of different techniques including dynamic time warping are compared. The word similarities are then used for clustering

It is almost universally accepted in speech recognition that phone- or word-level segmentation prior to recognition is neither feasible nor desirable, and in the dynamic (pen-based) handwriting recognition domain the success of segmentation-free techniques points to the same conclusion. But in image-based handwriting recognition, this conclusion is far from being firmly established, and the results presented in this paper show that systems employing character-level presegmentation can be more effective, even within the same HMM paradigm, than systems relying on sliding window feature extraction. We describe two variants of a Hidden Markov system recognizing handwritten addresses on US mail, one with presegmentation and one without, and report results on the CEDAR data set. 1. INTRODUCTION Any approach to speech and handwriting recognition must take into account that the signal is composed from a succession of alphabetic units (phonemes or graphemes). In the early work on speech recog...

We propose a statistical technique of zone finding for the class of documents that are neither rigidly structured like tax forms nor very unstructured like magazine pages or engineering drawings. Given an initial window assumed to contain the final zone (bounding box) of interest, and a `signature ' of the target, we propose to locate the final zone by a combination of simple outside in and inside out searches based on the assumption that the coordinates of the target have unimodal distribution. Results are presented in the bank check domain, and the applicability of the technique to other domains is discussed. 0. Introduction Proc 13th ICPR, Vienna (1996) Vol III, pp 818-822 Real world Optical Character Recognition (OCR) systems rarely enjoy the luxury, often taken for granted in more academic systems, of working with clearly delineated text zones. In fact, the task of zoning, or region extraction, i.e. identifying and precisely demarcating the zone(s) containing the text to be recog...

e author [2] under three conditions: using a bigram language model, a finite state grammar, and a combination of the two. As it is evident from Fig. 1, the rule-based system fares much better than the statistics-based, and in fact the latter adds very little to the performance of a system already containing the former. So the question is not so much an overarching philosophical problem of whether rules are better, but rather the more mundane practical problem of finding the rules. To the extent that the rules, constrains and representations constituting the grammar are devised by grammarians like Quirk and Greenbaum [3], computational linguists can get a free ride, and they should avail themselves of the opportunity. But to the extent that Quirk et al. represent the culmination of an extremely sophisticated descriptive tradition of a singularly deeply researched language, it appears very unlikely that more than a handful languages could be handled in the same fashion. Again as an ill