Multilayer neural networks trained with the back-propagation algorithm constitute the best example of a successful gradient based learning technique. Given an appropriate network architecture, gradient-based learning algorithms can be used to synthesize a complex decision surface that can classify high-dimensional patterns, such as handwritten characters, with minimal preprocessing. This paper reviews various methods applied to handwritten character recognition and compares them on a standard handwritten digit recognition task. Convolutional neural networks, which are specifically designed to deal with the variability of 2D shapes, are shown to outperform all other techniques. Real-life document recognition systems are composed of multiple modules including field extraction, segmentation recognition, and language modeling. A new learning paradigm, called graph transformer networks (GTN), allows such multimodule systems to be trained globally using gradient-based methods so as to minimize an overall performance measure. Two systems for online handwriting recognition are described. Experiments demonstrate the advantage of global training, and the flexibility of graph transformer networks. A graph transformer network for reading a bank cheque is also described. It uses convolutional neural network character recognizers combined with global training techniques to provide record accuracy on business and personal cheques. It is deployed commercially and reads several million cheques per day.

Gradient-based learning applied to document recognition

Machine Learning is the study of methods for programming computers to learn. Computers are applied to a wide range of tasks, and for most of these it is relatively easy for programmers to design and implement the necessary software. However, there are many tasks for which this is difficult or impossible. These can be divided into four general categories. First, there are problems for which there exist no human experts. For example, in modern automated manufacturing facilities, there is a need to predict machine failures before they occur by analyzing sensor readings. Because the machines are new, there are no human experts who can be interviewed by a programmer to provide the knowledge necessary to build a computer system. A machine learning system can study recorded data and subsequent machine failures and learn prediction rules. Second, there are problems where human experts exist, but where they are unable to explain their expertise. This is the case in many perceptual tasks, such as speech recognition, hand-writing recognition, and natural language understanding. Virtually all humans exhibit expert-level abilities on these tasks, but none of them can describe the detailed steps that they follow as they perform them. Fortunately, humans can provide machines with examples of the inputs and correct outputs for these tasks, so machine learning algorithms can learn to map the inputs to the outputs. Third, there are problems where phenomena are changing rapidly. In finance, for example, people would like to predict the future behavior of the stock market, of consumer purchases, or of exchange rates. These behaviors change frequently, so that even if a programmer could construct a good predictive computer program, it would need to be rewritten frequently. A learning program can relieve the programmer of this burden by constantly modifying and tuning a set of learned prediction rules. Fourth, there are applications that need to be customized for each computer user separately. Consider, for example, a program to filter unwanted electronic mail messages. Different users will need different filters. It is unreasonable to expect each user to program his or her own rules, and it is infeasible to provide every user with a software engineer to keep the rules up-to-date. A machine learning system can learn which mail messages the user rejects and maintain the filtering rules automatically. Machine learning addresses many of the same research questions as the fields of statistics, data mining, and psychology, but with differences of emphasis. Statistics focuses on understanding the phenomena that have generated the data, often with the goal of testing different hypotheses about those phenomena. Data mining seeks to find patterns in the data that are understandable by people. Psychological studies of human learning aspire to understand the mechanisms underlying the various learning behaviors exhibited by people (concept learning, skill acquisition, strategy change, etc.).

Machine learning

The primary goal of pattern recognition is supervised or unsupervised classification. Among the various frameworks in which pattern recognition has been traditionally formulated, the statistical approach has been most intensively studied and used in practice. More recently, neural network techniques and methods imported from statistical learning theory have been receiving increasing attention. The design of a recognition system requires careful attention to the following issues: definition of pattern classes, sensing environment, pattern representation, feature extraction and selection, cluster analysis, classifier design and learning, selection of training and test samples, and performance evaluation. In spite of almost 50 years of research and development in this field, the general problem of recognizing complex patterns with arbitrary orientation, location, and scale remains unsolved. New and emerging applications, such as data mining, web searching, retrieval of multimedia data, face recognition, and cursive handwriting recognition, require robust and efficient pattern recognition techniques. The objective of this review paper is to summarize and compare some of the well-known methods used in various stages of a pattern recognition system and identify research topics and applications which are at the forefront of this exciting and challenging field.

/pdf/statistical-pattern-recognition-a-review-1kz1b0tte4.pdf

Statistical pattern recognition: a review

We develop a common theoretical framework for combining classifiers which use distinct pattern representations and show that many existing schemes can be considered as special cases of compound classification where all the pattern representations are used jointly to make a decision. An experimental comparison of various classifier combination schemes demonstrates that the combination rule developed under the most restrictive assumptions-the sum rule-outperforms other classifier combinations schemes. A sensitivity analysis of the various schemes to estimation errors is carried out to show that this finding can be justified theoretically.

On combining classifiers

From the Publisher:
Pattern recognition has long been studied in relation to many different (and mainly unrelated) applications, such as remote sensing, computer vision, space research, and medical imaging. In this book Professor Ripley brings together two crucial ideas in pattern recognition; statistical methods and machine learning via neural networks. Unifying principles are brought to the fore, and the author gives an overview of the state of the subject. Many examples are included to illustrate real problems in pattern recognition and how to overcome them.This is a self-contained account, ideal both as an introduction for non-specialists readers, and also as a handbook for the more expert reader.

Pattern recognition and neural networks

In this paper, we initiate a theoretical study on ${N}$ -expert fusion ( ${N}~\boldsymbol {\geq }~2$ ) in the context of biometric authentication (BA). Optimal fusion weights, which depend on performances and variances of, and correlations among individual base-experts have been found, and we also give and prove some new theorems that serve as the basis for analyzing the performance of the overall system. Our conclusion is that provided that optimal weights are used as fusion coefficients, linear fusion will definitely lead to a better performance than the best individual expert. This contradicts many existing conclusions, which assert that fusion is not always beneficial and that performance improvement due to fusion is guaranteed only when some conditions as to base-experts’ performances, variances, and correlations are satisfied. Besides, for the first time the definition of correlation in the context of BA is clearly and explicitly given to avoid the long-standing ambiguity and vagueness concerning this term, and we make an initial attempt to propose and investigate three types of correlation coefficients. Furthermore, the connection between our proposed optimal fusion method and Fisher’s discriminant is discussed. Extensive experiments have been conducted to confirm our theoretical results and construct counter-examples for the existing conclusions.

A Study on Performance Improvement Due to Linear Fusion in Biometric Authentication Tasks

Robust handwriting recognition of complex patterns of arbitrary scale, orientation and location is yet elusive to date as reaching a good recognition rate is not trivial for most of the application developments in this field. Cursive scripts with complex character shapes, such as Arabic and Persian, make the recognition task even more challenging. This complexity requires sophisticated representations and learning methods, and comprehensive data samples. A direct approaches to achieve a better performance is focusing on designing more powerful building blocks of a handwriting recognition system which are pattern representation and pattern classification . In this paper we aim to scale up the efficiency of online recognition systems for Arabic characters by integrating novel representation techniques into efficient classification methods. We investigate the idea of incorporating two novel feature representations for online character data. We advocate the usefulness and practicality of these features in classification methods using neural networks and support vector machines. The combinations of proposed representations with related classifiers can offer a module for recognition tasks which can deal with any two-dimensional online pattern. Our empirical results confirm the higher distinctiveness and robustness to character deformations obtained by the proposed representation compared to currently available techniques.

/pdf/integration-of-contextual-information-in-online-handwriting-32o470x5uh.pdf

Integration of Contextual Information in Online Handwriting Representation

HMM has been successfully used to model 1D data, e.g. voice signals. Their use to model 2D patterns was not as successful due to a major difficulty, in describing the 2D data using 1D observation sequences. In this paper, we discuss the importance of this issue and present an improved method to extract 1D observations from the dynamics of off-line handwritten words. The method is based on pen trajectory estimation techniques. The paper also includes description of our HMM classifier which allows dynamic termination states to achieve enhanced discriminative power. Experimental results show the applicability and usefulness of the proposed method. As a result of using the termination probability in HMM modeling, the top 1/sup st/ recognition rate increased by 10%.

Dynamic observations and dynamic state termination for off-line handwritten word recognition using HMM

On the Recognition of Handwritten Characters using Neural Networks.

Dans cet article nous introduisons une nouvelle strategie permettant l'amelioration d'un jeu de primitives d'un systeme de reconnaissance de l'ecriture manuscrite. L'idee de base repose sur la constatation que parmi un ensemble de primitives, certaines sont plus discriminantes que d'autres. La technique d'amelioration proposee consiste a conserver les primitives les plus discriminantes et a substituer celles qui ne le sont pas. Dans un premier temps le pouvoir discriminant de chaque primitive doit etre evalue, la perplexite conditionnelle est utilisee a cet effet. Ensuite l'identification de classes de primitives non-discriminantes permet de choisir la strategie de descente d'un niveau perceptif. Un nouvel ensemble de primitives specialise pour chaque sous-probleme de classification identifie par ces classes de primitives est ensuite defini. Leur regroupement avec les primitives discriminantes initiales permet de construire un nouveau systeme de reconnaissance ameliore. L'application de cette technique sur un systeme existant a montre sa pertinence.

/pdf/une-nouvelle-strategie-pour-l-amelioration-des-jeux-de-4dn4cgsabx.pdf

Ching Y. Suen

Papers

A Study on Performance Improvement Due to Linear Fusion in Biometric Authentication Tasks

Integration of Contextual Information in Online Handwriting Representation

Dynamic observations and dynamic state termination for off-line handwritten word recognition using HMM

On the Recognition of Handwritten Characters using Neural Networks.

Une nouvelle stratégie pour l'amélioration des jeux de primitives d'un système de reconnaissance de l'écriture