General Pattern Recognition Research Articles

A new set of constraint-free character recognition grammars

A general pattern-recognition procedure for application to unconstrained alphanumeric characters is presented. The procedure is designed to allow hierarchical redescription of the input images in terms of significant elements only, and formal developments are given within the framework of elementary phrase-structure grammars. The extraction of the primitives associated with the terminal vocabularies of the grammars is always deterministic, and the productions of the parsers are characterized by a significant degree of topological fidelity. Preliminary experimental results indicate recognition rates comparable to the state of the art, but with a considerable reduction in computing time.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Recognition of general patterns using neural networks

The Hopfield model of neural network stores memory in its symmetric synaptic connections and can only learn to recognize sets of nearly "orthogonal" patterns. A new algorithm is put forth to permit the recognition of general ("non-orthogonal") patterns. The algorithm specifies the construction of the new network's memory matrix Tij, which is, in general, asymmetrical and contains the Hopfield neural network (Hopfield 1982) as a special case. We find further that in addition to this new algorithm for general pattern recognition, there exists in fact a large class of Tij memory matrices which permit the recognition of non-orthogonal patterns. The general form of this class of Tij memory matrix is presented, and the projection matrix neural network (Personnaz et al. 1985) is found as a special case of this general form. This general form of memory matrix extends the library of memory matrices which allow a neural network to recognize non-orthogonal patterns. A neural network which followed this general form of memory matrix was modeled on a computer and successfully recognized a set of non-orthogonal patterns. The new network also showed a tolerance for altered and incomplete data. Through this new method, general patterns may be taught to the neural network.

Learning with mislabeled training samples using stochastic approximation

For the problem of parameter learning in pattern recognition, the convergence of stochastic approximation-based learning algorithms have been investigated for the situation in which mislabeled training samples are present. In the cases considered, it is found that estimates converge to nontrue values in the presence of labeling errors. The general m-class N-feature pattern recognition problem is considered. A possible solution to the problem is also discussed. Some simulation results are provided to support the conclusions drawn.

Methodologies in pattern recognition and image analysis—A brief survey

This paper presents an overview of the current methodologies on image analysis and pattern recognition. Both subjects are better known under the general term image pattern recognition. At first a distinction is made between those two streams and later each subject is treated from a methodological point of view. Image analysis is described as that field of science that studies image descriptions. Major areas of image analysis methodologies are matching, segmentation, shape analysis and description. On the other hand, pattern recognition studies the classification processes and has as its main streams of research the statistical, the syntactic and finally hybrid methods of the previous two approaches. Some future tendencies are also explained.

Least desirable feature elimination in a general pattern recognition problem

A new technique for dimensionality reduction for a general pattern recognition problem with pattern classes represented by multivariate normal distributions is presented. The method consists of identifying and eliminating the least desirable feature out of the original feature space. It is shown that great simplicity is obtained by doing so in comparison to the existing methods. A very simple expression describing a vector representing the least desirable feature in a most general case is derived. If the original feature space is N-dimensional then recognizing and eliminating such a feature is equivalent to selecting and retaining the N−1 best features. J-divergence is used as a measure of the discrimination between the classes. The flow chart for the method for discarding more than one dimension is presented.

Turn prediction in proteins using a pattern-matching approach.

We extend the use of amino acid sequence patterns [Cohen, F.E., Abarbanel, R. M., Kuntz, I. D., & Fletterick, R. J. (1983) Biochemistry 22, 4894-4904] to the identification of turns in globular proteins. The approach uses a conservative strategy, combined with a hierarchical search (strongest patterns first) and length-dependent masking, to achieve high accuracy (95%) on a test set of proteins of known structure. Applying the same procedure to homologous families gives a 90% success rate. Straightforward changes are suggested to improve the predictive power. The computer program, written in Lisp, provides a general pattern-recognition language well suited for a number of investigations of protein and nucleic acid sequences.

Optimizing pulse pileup detection for soft X-ray spectroscopy

The problem of optimizing detection of the pileup of randomly occuring exponential tail pulses in white noise is considered. An attempt is made to reduce the process to an algorithm that could practically be performed in real time. Quantitative estimates are made for the performance of such an “optimum” detector. The relation to a more general pattern recognition problem is mentioned.

The contour extraction problem with biomedical applications

The problem of extracting a contour of some sort from noisy digitized images often arises within the context of the general pattern recognition problem. The approach presented here formulates the contour extraction problem as one of minimum cost tree searching. Branch costs or metrics are defined which are indicative of the likelihood that a particular branch lies on the true contour. The branch metrics incorporate both local and global or contextual information. The most likely path or contour is then extracted by application of a heuristic tree searching algorithm. In particular, use is made of the Zigangirov-Jelinek (ZJ) stack algorithm. This algorithm avoids the computational burden associated with structured or exhaustive search techniques and results in more manageable although random computational requirements. The approach is described first on an artificially contrived problem and then applied to two real biomedical image processing problems. The biomedical problems considered are concerned with the extraction of the left ventricular outline from serial angiocardiograms and the extraction of lung outlines from digitized chest X rays. It is demonstrated that the approach is fast, reliable, and versatile in the sense that it can be applied to a wide variety of contour extraction problems by appropriate definition of branch metrics.

A powerful and efficient structural pattern recognition system

This paper describes a general structural pattern recognition system. It introduces code—a description language for patterns (also called concepts) based on boolean logic and set theory. Internal machine representations for concepts and objects are given which directly influence the efficiency of the system. In particular, a data structure, called a “graft”, is described. A method of storing and accessing concept descriptions is also given. This feature of the system is partly an attempt to restrict the recognition process to those concepts that might be relevant to an object. Finally, the recognition process is described and the reasons for its power and efficiency are discussed.

Computational geometry of linear threshold functions

Linear threshold machines are defined to be those whose computations are based on the outputs of a set of linear threshold decision elements. The number of such elements is called the rank of the machine. An analysis of the computational geometry of finite-rank linear threshold machines, analogous to the analysis of finite-order perceptrons given by Minsky and Papert, reveals that the use of such machines as “general purpose pattern recognition systems” is severely limited. For example, these machines cannot recognize any topological invariant, nor can they recognize nontrivial figures “in context”.

Heuristics for Signature Table Analysis as a Pattern Recognition Technique

The signature table method is a hierarchical approach for the recognition of binary patterns which are described by means of many features. The method was first applied by A. L. Samuel and his students for prediction of whether a given checker position was master quality or not. The variants of the methods discussed herein consider it as a general pattern recognition technique. Examples are presented to suggest both the potential and hazards of using the method for recognition of patterns. Advantages of the method are the extraction of nonlinear Boolean relationships among the variables, the use of incomplete data in a routine way, and that it sometimes may provide better prediction at less cost than multiple regression. The principle disadvantage, the lack of a theory to guide the choice of key parameters in the method, can sometimes be overcome by systematic computer search.

A class of error tolerant pattern discrimination functions

A general pattern recognition problem is posed in Hilbert space. Two new solutions are then given and it is shown that the sensitivity of the pattern recognition functions to pattern perturbation can be a priori controlled. A series of examples demonstrate the principal results in a variety of settings.

Estimation of feature-dependent markov process transition probability matrices

This article is concerned with the estimation of Markov process transition probabilities for nonhomogeneous populations. This estimation problem is shown to be solvable using general pattern recognition techniques. Numerous multivariate estimation techniques exist in the field of statistical pattern recognition, and many of these will be useful to researchers who use Markov process models of a population's behavior. These techniques are particularly called for when the behavior of members of a population is suspected to depend upon a set of descriptive feature values. One such technique, that of linear discriminant analysis, is presented in detail as an illustration of a statistical pattern recognition approach to a specific Markov process estimation problem. A brief example is given.

Geophysical interpretation — A pattern recognition system

The general interpretation system is defined by a cyclic chain of deductions and inductions within the premises of the general pattern recognition system. The major problem of adducing the best-fitting model for the determination of target parameters through their field response can be achieved by designing an adaptive pattern recognition scheme. Such a scheme has been resolved into three major areas: Prognosis, Sophistication and Diagnosis. Sophistication has been considered as the key to simulate natural settings. The dependence of this sophistication on the type of physical parameter involved has been shown through statistical approach of Kernel approximation.

Some Upper Bounds on Error Probability for Multiclass Pattern Recognition

An upper bound on the probability of error for the general pattern recognition problem is obtained as a functional of the pairwise Kolmogorov variational distances. Evaluation of the bound requires knowledge of a priori probabilities and of the class-conditional probability density functions. A tighter bound is obtained for the case of equal a priori probabilities, and a further bound is obtained that is independent of the a priori probabilities.

Probability of Error Bounds

Abstract : Simplified upper and lower bounds to the probability of error for general M-ary hypotheses pattern recognition are obtained. The bounds, applicable to general non-gaussian densities and especially mixture densities encountered in adaptive pattern recognition, are simple to calculate and hence valuable for on-line performance evaluation of pattern recognition system. Computer evaluation of the bounds, established their tight nature and computational simplicity. Based on the bounds, feature extraction criteria are derived for supervised as well as parametric adaptive pattern recognition. (Author)

A class of upper bounds on probability of error for multihypotheses pattern recognition (Corresp.)

A class of upper bounds on the probability of error for the general multihypotheses pattern recognition problem is obtained. In particular, an upper bound in the class is shown to be a linear functional of the pairwise Bhattacharya coefficients. Evaluation of the bounds requires knowledge of a priori probabilities and of the hypothesis-conditional probability density functions. A further bound is obtained that is independent of a priori probabilities. For the case of unknown a priori probabilities and conditional probability densities, an estimate of the latter upper bound is derived using a sequence of classified samples and Kernel functions to estimate the unknown densities.

The pupillary control system: Its non-linear adaptive and stochastic engineering design characteristics

In the decade since the pupillary light reflex to light was first studied using a linear transfer function approach [18, 20], many developments in control theory and in research into biological control systems have occurred. Optimal control theory, Liapunov stability methods, Wiener-G functional analysis, general systems identification procedures and pattern recognition techniques for adaptive systems have influenced experimental approaches to a wide variety of neurological control systems. An intensive reinvestigation using classical electrophysiological techniques to define interesting design features of the underlying neurophysiological mechanisms of these systems is now proceeding, and it appears that clinical applications are on the horizon. Three contemporary experimental approaches to the pupillary system will be described in detail and shown to result in complementary views. 1. (A) Wiener-G functional analysis using both random and time domain inputs [10]. 2. (B) Statistical communication analysis of pupil noise [12, 16], and 3. (C) Neurophysiological techniques embedded within a systems approach [19].

Characterization of an adaptive filter for the analysis of variable latency neuroelectric signals

An adaptive filter system has been developed whereby variable latency neuroelectric signals may be detected and separated from associated noise. The system is based on correlation-averaging techniques which are described in detail. The adaptive property of the system derives from iterative correlation and averaging of the data signals and permits the recognition of signals the specific waveshapes of which are not known in advance. The system consitutes a general pattern recognition device. It has been shown to be applicable to the analysis of evoked potentials of variable latency as well as to the analysis of patterns of EEG activity. Further applications include the analysis of multiple, complex signals such as miniature potentials from single motor units in the spinal cord. The use of an adaptive filter such as this with convergence properties based on broad statistical considerations appears to have greater analytic power than have previous methods.

Hilbert space methods for detection theory and pattern recognition

The problem of classifying an observation into one of several different categories, or patterns, is considered. The observation consists of a sample function of a continuous-time parameter stochastic process observed over a finite-time interval. When only two categories are involved the general pattern recognition problem reduces to the signal detection problem. The methods used are based upon results from the theory of reproducing kernel Hilbert spaces. This theory has been developed within the last few years and the application of these results to problems of statistical inference for stochastic processes has taken place only recently. Therefore, a reasonably serf-contained exposition of the results required from the theory of reproducing kernel Hilbert spaces is presented. It is pointed out that the decision rule employed by the optimum pattern recognition system is based on the likelihood ratio. This quantity exists fi, and only if, the probability measures are equivalent, i.e., mutually absolutely continuous with respect to each other. In the present work only Gaussian processes are considered, in which case it is well known that the probability measures can only be either equivalent or perpendicular, i.e., mutually singular. It is shown that the reproducing kernel Hilbert space provides a natural tool for investigating the equivalence of Gaussian measures. In addition, this approach provides a convenient means for actually evaluating the likelihood ratio. The results are applied to two pattern recognition problems. The first problem involves processes which have the same covariance function but different mean-value functions and the second problem concerns processes with different covariance functions and zero mean-value functions.