Despite the success of many pattern recognition systems in constrained domains, problems that involve noisy input and many classes remain difficult. A promising direction is to use several classifiers simultaneously, such that they can complement each other in correctness. This thesis is concerned with decision combination in a multiple classifier system that is critical to its success. A multiple classifier system consists of a set of classifiers and a decision combination function. It is a preferred solution to a complex recognition problem because it allows simultaneous use of feature descriptors of many types, corresponding measures of similarity, and many classification procedures. It also allows dynamic selection, so that classifiers adapted to inputs of a particular type may be applied only when those inputs are encountered. Decisions by the classifiers are represented as rankings of the class set that are derivable from the results of feature matching. Rank scores contain more ...

Recognition of degraded text is a challenging problem. To improve the performance of an OCR system on degraded images of text, postprocessing techniques are critical. The objective of postprocessing is to correct errors or to resolve ambiguities in OCR results by using contextual information. Depending on the extent of context used, there are different levels of postprocessing. In current commercial OCR systems, word-level postprocessing methods, such as dictionary-lookup, have been applied successfully. However, many OCR errors cannot be corrected by word-level postprocessing. To overcome this limitation, passage-level postprocessing, in which global contextual information is utilized, is necessary. In most current studies on passage-level postprocessing, linguistic context is the major resource to be exploited. This thesis addresses problems in degraded text recognition and discusses potential solutions through passage-level postprocessing. The objective is to develop a postprocessin...

The Levenshtein-distance between two words is the minimal number of insertions, deletions or substitutions that are needed to transform one word into the other. Levenshtein-automata of degree n for a word W are defined as finite state automata that regognize the set of all words V where the Levenshtein-distance between V and W does not exceed n. We show how to compute, for any fixed bound n and any input word W , a deterministic Levenshtein-automaton of degree n for W in time linear in the length of W . Given an electronic dictionary that is implemented in the form of a trie or a finite state automaton, the Levenshtein-automaton for W can be used to control search in the lexicon in such a way that exactly the lexical words V are generated where the Levenshtein-distance between V and W does not exceed the given bound. This leads to a very fast method for correcting corrupted input words of unrestricted text using large electronic dictionaries. We then introduce a second method that avoids the explicit computation of Levenshtein-automata and leads to even improved eciency. We also describe how to extend both methods to variants of the Levenshtein-distance where further primitive edit operations (transpositions, merges and splits) may be used.

Contents 1 Background 4 1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 1.2 Types of handwriting recognition systems . . . . . . . . . . . . . . . . 5 1.2.1 Ooe-line recognition . . . . . . . . . . . . . . . . . . . . . . . . 5 1.2.2 On-line recognition . . . . . . . . . . . . . . . . . . . . . . . . 5 1.2.3 Character sets . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 1.2.4 Writing style variations . . . . . . . . . . . . . . . . . . . . . . 7 1.2.4.1 Variations of characters . . . . . . . . . . . . . . . . 7 1.2.4.2 Alignment of characters . . . . . . . . . . . . . . . . 8 1.2.4.3 Personal background factors . . . . . . . . . . . . . . 8 1.2.4.4 Situational factors . . . . . . . . . . . . . . . . . . . 9 1.2.4.5 Material factors . . . . . . . . . . . . . . . . . . . . . 9 1.2.4.6 Constraints on writing . . . . . . . . . . . . . . . . . 9 1.2.5 Context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 1.3 Beneøt

The need to correct garbled strings arises in many areas of natural language processing. If a dictionary is available that covers all possible input tokens, a natural set of candidates for correcting an erroneous input P is the set of all words in the dictionary for which the Levenshtein distance to P does not exceed a given (small) bound k. In this article we describe methods for efficiently selecting such candidate sets. After introducing as a starting point a basic correction method based on the concept of a "universal Levenshtein automaton," we show how two filtering methods known from the field of approximate text search can be used to improve the basic procedure in a significant way. The first method, which uses standard dictionaries plus dictionaries with reversed words, leads to very short correction times for most classes of input strings. Our evaluation results demonstrate that correction times for fixed-distance bounds depend on the expected number of correction candidates, which decreases for longer input words. Similarly the choice of an optimal filtering method depends on the length of the input words.

Given a bitmapped image of a page from any document, a page-reading system identifies the characters on the page and stores them in a text file. This “OCR-generated” text is represented by a string and com-pared with the correct string to determine the accuracy of this process. The string editing problem is applied to find an optimal correspondence of these strings using an appropriate cost function. The ISRI annual test of page-reading systems utilizes the following performance measures, which are defined in terms of this correspondence and the string edit distance: character accuracy, throughput, accuracy by character class, marked char-acter efficiency, word accuracy, non-stopword accuracy, and phrase accu-racy. It is shown that the universe of cost functions is divided into equivalence classes, and the cost functions related to the longest common subsequence (LCS) are identified. The computation of a LCS can be made faster by a linear-time preprocessing step.

The research presented in this thesis provides new solutions to two fundamental problems in document recognition. The first problem is character segmentation. Touching characters and character fragmentation caused by image degradation are difficult problems for current document recognition systems. The second problem that this thesis addresses is the dependence of today's document recognition systems on extensive font training. These two problems are shown to be the main reasons that cause performance breakdown for most of today's commercial document recognition systems. Our research provides solutions to the two problems by seeking alternative approaches that can recognize degraded documents with high accuracy and robust performance. Reliable performance on degraded documents is achieved by avoiding these two difficult problems in the recognition approaches. We propose to consider the computational task of document recognition as a process of finding the mapping between the visual pat...

In this paper a new algorithm for string edit distance computation is proposed. It is based on the classical approach [11]. However, while in [11] the two strings to be compared may be given online, our algorithm assumes that one of the two strings to be compared is a dictionary entry that is known a priori. This dictionary word is converted, in an off-line phase to be carried out beforehand, into a special type of deterministic finite state automaton. Now, given an input string corresponding to a word with possible OCR errors and the automaton derived from the dictionary word, the computation of the edit distance between the two strings corresponds to a traversal of the states of the automaton. This procedure needs time which is only linear in the length of the OCR word. It is independent of the length of the dictionary word. Given not only one but N different dictionary words, their corresponding automata can be combined into a single deterministic finite state automaton. Thus the co...

A document image is a visual representation of a printed page such as a journal article page, a facsimile cover page, a technical document, an o#ce letter, etc. Document understanding as a research endeavor consists of studying all processes involved in taking a document through various representations: from a scanned physical document to high-level semantic descriptions of the document. Some of the types of representation that are useful are: editable descriptions, descriptions that enable exact reproductions and high-level semantic descriptions about document content. This report is a de#nition of #ve research subdomains within document understanding as pertaining to predominantly printed documents. The topics described are: modular architectures for document understanding; decomposition and structural analysis of documents; model-based OCR; table, diagram and image understanding; and performance evaluation under distortion and noise. 1 Each of the main sections of this paper were ...

In this paper we present a foundational basis for optimal and information theoretic syntactic pattern recognition. We do this by developing a rigorous model, M * , for channels which permit arbitrarily distributed substitution, deletion and insertion syntactic errors. More explicitly, if A is any finite alphabet and A * the set of words over A, we specify a stochastically consistent scheme by which a string U A * can be transformed into any Y A * by means of arbitrarily distributed substitution, deletion and insertion operations. The scheme is shown to be Functionally Complete and stochastically consistent. Apart from the synthesis aspects, we also deal with the analysis of such a model and derive a technique by which Pr[Y|U], the probability of receiving Y given that U was transmitted, can be computed in cubic time using dynamic programming. One of the salient features of this scheme is that it demonstrates how dynamic programming can be applied to evaluate quantities involv...