Handwriting recognition systems based on hidden Markov models commonly use a vector quantizer to get the required symbol sequence. In order to get better recognition rates semi-continuous hidden Markov models have been applied. Those recognizers need a soft vector quantizer which superimposes a statistical distribution for symbol generation. In general, Gaussian distributions are applied. A disadvantage of this technique is the assumption of a specific distribution. No proof can be given whether this presupposition holds in practice. Therefore, the application of a method which employs no model of a distribution may achieve some improvements. The paper presents the employment of a polynomial classifier as a replacement of a Gaussian classifier in the handwriting recognition system. The replacement improves the recognition rate significantly, as the results show.

The most scientific papers dealing with handwriting recognition systems make statements relating to the recognition performance based on a forced-recognition rate. This rate describes the ratio between the number of the correct recognized samples and the number of all possible samples. For systems applied in real applications this rate is not very relevant. They have to work with a very low error-rate, which can be only achieved by introducing effective reject criteria. So the real interesting thing is a function describing the recognition rate in relation to a specific error rate, including implicitly a corresponding reject rate. This paper presents two approaches for handling rejects in a hidden Markov based handwriting recognition system. The features to determine a reject are values which are derived from the hidden Markov recognizer. One of the techniques relies on relative frequencies of those values, the other one utilizes standard classification techniques to train a reject decision unit, the reject classifier. Both methods are presented with some noteworthy results.

In a binary image contours may be seen as the discriminating curve between objects and background. Contours of connected components are always a Jordan curve. One symbol (e.g., a character) may consist of more than one such curve. Processing these curves is a one-dimensional task. Almost all common processing steps can be designed to work on contours rather than on the two-dimensional image. Moreover, contour processing gives new insight to well know problems and enables new processing steps or produces more information about the relations between connected components or objects of the image. The authors present preprocessing operations which work directly on the level of contours. Compared to the corresponding iconic operations, algorithms working on the contour level are mostly more efficient. Based on the contours of the connected components methods for filtering and slant normalization are described.

Estimation of aiding typographical rulers is a challenging task especially for everyday handwriting. The method described here performs very well as long as the model assumption of one straight line is satisfied. It is completely based on contour processing. A sophisticated iterated regression analysis processing weighted points is the central algorithm. The method can be used for handwriting as well as machine printed texts without adjusting the parameters.

Handwriting recognition systems usually need the support of lexical knowledge in order to achieve acceptable results. Lexicons of practical applications are often very large which results in prohibitive run time and recognition performance. So there is a need to reduce large lexicons efficiently without loosing the correct entry. Often it is possible to recognize some isolated or resegmented characters of a word but not the whole word. These recognition results may be used as hints for an initial lexicon reduction. In order to use these hints techniques are needed which are able to handle character alternatives as well as touched and broken characters. The article discusses lexicon techniques in respect to their efficiency and robustness. A hybrid approach is proposed which reduces large lexicons efficiently and shows a robust behavior when broken and touched characters are observed.

Offline cursive script word recognition has received increasing attention during the last years. Impressive progress has been achieved in reading isolated single characters during the last decade. Cursive script recognition still lacks a good recognition rate. Since there is a high variability in unconstrainted handwritten script words, the domain is much more difficult than single character recognition. To achieve acceptable results, the context has to be restricted by a given lexicon of all possible words. The only accessible information is the binary image of the cursive script word. Since handling of raster data is cumbersome, connectivity analysis is applied as a first processing step. Thereafter it is necessary to reduce the variability as much as possible without losing relevant information. Therefore, some normalization steps angle, rotation stroke width, and size. The normalization techniques of the authors' system and the subsequent feature extraction are presented. The proposed algorithms are every efficient because they are based on the contour information provided by connectivity analysis.>

A methodology for structuring large disordered sample sets for classifiers is presented. The object-oriented framework is an essential part of this methodology. Classes can be viewed as sets, and sets again can be viewed as objects. For this reason, operations and techniques from both domains (sets and OO technology) can be utilized to set up a system for computer-aided labeling. Since labeling is a time-consuming task, the handling of the system has to support efficient labeling. A second important aspect of the application is easy system handling to allow inexperienced examiners to use the system.>

The paper describes an adaptation of hidden Markov models (HMM) to automatic recognition of unrestricted handwritten words. Many interesting details of a 50,000 vocabulary recognition system for US city names are described. This system includes feature extraction, classification, estimation of model parameters, and word recognition. The feature extraction module transforms a binary image to a sequence of feature vectors. The classification module consists of a transformation based on linear discriminant analysis and Gaussian soft-decision vector quantizers which transform feature vectors into sets of symbols and associated likelihoods. Symbols and likelihoods form the input to both HMM training and recognition. HMM training performed in several successive steps requires only a small amount of gestalt labeled data on the level of characters for initialization. HMM recognition based on the Viterbi algorithm runs on subsets of the whole vocabulary.>

Two different approaches are described to combine the results of different classifiers. The first approach is based on the Dempster/Shafer theory of evidence and the second one is a statistical approach with some assumptions on the input data. Both approaches were tested for user-dependent recognition of on-line handwritten characters.>

The authors present a conceptual framework for solving the task of document analysis, which, in essence, consists in the conversion of the document's pixel representation into an equivalent knowledge network representation holding the document's content and layout. Starting on the pixel level, the formation of elementary geometric objects on which layout analysis as well as the definition of character objects is based is described. Character recognition accomplishes the mapping from geometric object to character meaning in ASCII representation. On the next level of abstraction words are formed and verified by contextual processing. Modeled knowledge about complete documents and about how their constituents are related to the application forms the highest level of abstraction. The various problems arising at each stage are discussed. The dependencies between the different levels are exemplified and technical solutions put forward.>

A method of representing connected components in multivalued images is proposed, and an efficient, sequential, one-pass algorithm for generating the border-line chain code for each component is introduced. Simultaneously, the algorithm produces for each component a list of all neighboring components that are encountered during a walk around the border line, thus providing full topological information. The number of different pixel values does not affect performance.>