Showing papers on "Feature (machine learning) published in 1974"

Pattern recognition principles

Abstract: The present work gives an account of basic principles and available techniques for the analysis and design of pattern processing and recognition systems. Areas covered include decision functions, pattern classification by distance functions, pattern classification by likelihood functions, the perceptron and the potential function approaches to trainable pattern classifiers, statistical approach to trainable classifiers, pattern preprocessing and feature selection, and syntactic pattern recognition.

Patterns in pattern recognition: 1968-1974

Abstract: This paper selectively surveys contributions to major topics in pattern recognition since 1968. Representative books and surveys pattern recognition published during this period are listed. Theoretical models for automatic pattern recognition are contrasted with practical,, design methodology. Research contributions to statistical and structural pattern recognition are selectively discussed, including contributions to error estimation and the experimental design of pattern classifiers. The survey concludes with a representative set of applications of pattern recognition technology.

A Theory of Character Recognition by Pattern Matching Method

Abstract: Character recognition is a categorizing process of unknown input pattern into one of known finite number of character categories. Various practical methods of realizing this process has been devised. Pattern matching method is one of the most commonly used techniques in which the similarity of input pattern is tested with the reference pattern of each category.

Speech recognition device for controlling a machine

Abstract: In an apparatus for speech recognition, intended for controlling machines, spoken words are synthesized from two to twelve phoneme classes. Phonemes are recognized by analysis of speech sounds, measuring and comparing energy and time-rate of change of energy in certain frequency bands. Words are recognized by further logic means analyzing phoneme classes. One particular feature determines the plosive class "T" versus the fricative class "S" by measuring energy rise-time.

Computerized pattern recognition applications to chemical analysis. Development of interactive feature selection methods for the K-nearest neighbor technique

Abstract: : A systematic approach has been developed for feature selection in the application of the K-nearest neighbor (KNN) computerized pattern recognition method. The approach uses an operator-interactive computer system. A large number of potentially-useful features for classification of patterns can be screened for the most relevant members by a combination of recommended procedures. These include: (a) one-dimensional KNN classification of all patterns using each feature individually; (b) inspection of histogram displays of classification records for each feature; and (c) establishment of consensus classifications from combined one-dimensional results. A computerized trial-and-error procedure can then be implemented to find the best combination of a minimum number of features for accurate classification using the multi-dimensional KNN method. (Modified author abstract)

Automatic target screening

Abstract: A feature extraction and selection technique for the recognition of chararistics identified with man-made objects within a scene of natural terrain, wherein the frequency of occurrence of the features is plotted in the form of three-dimensional histograms which describe the features of man-made objects, such as straight edges and regular geometric shapes. Employing conventional pattern recognition techniques, these features are used to classify the imagery as man-made or non man-made.

Computer Identification of Constrained Handprinted Characters with a High Recognition Rate

Abstract: Handprinted character recognition by computer is accomplished on forty-nine character classes with a high recognition rate (> 99.4 percent). The form of characters is constrained by requiring each character to be handprinted on a standard grid. The grid is composed of twenty line segments, each of which forms the basis for a feature, yielding twenty features to represent each character. A person printing these characters is not expected to remain precisely on the grid lines. The errors that do occur in following the grid lines are assumed to be normally distributed; therefore, each feature is based on a "longitudinal Gaussian-shaped surface." A page of constrained characters to be recognized is input to the computer using a television camera. Each character on the page is located, isolated from the other characters, and quantized into binary points. A novel technique utilizing a Gaussian point-to-line distance concept for calculation of "feature value" has been employed. The recognition program extracts the twenty feature values and attempts to determine in which of the forty-nine character classes the unknown character belongs. This decision is made based on these twenty features using a weighted minimum distance classifier. If only a marginal classification can be made, a second-level decision is used to increase the likelihood of correct classification. The second-level decision uses the most discriminating features of the two most likely character classes in order to increase the likelihood of correct classification. All character-dependent data are obtained through training techniques.

Some New Error Bounds and Approximations for Pattern Recognition

Abstract: For the average error probability Pe associated with the Bayes recognition procedures for two possible patterns, using no context, new upper and lower bounds and approximations are obtained. Results are given in terms of simple functions of feature "reliability" and a priori probabilities of the patterns. Two kinds of feature "reliability" are considered, i.e., distance between probability distributions and error probabilities without the use of a priori probabilities. Computational advantages offered by those bounds and approximations are pointed out. The question as to how close they are to P e is examined. In some special cases, they are perfect. Numerical examples show that the differences are in general about 5-10 percent, and comparisons with certain known results are quite favorable. Possible applications are discussed. Extension is also made to m possible patterns arranged in a hierarchy with two elements at each branching.

Special feature: An interactive minicomputer laboratory for graphics, image processing, and pattern recognition

Abstract: An interactive computer environment is one which attempts to facilitate the interplay between man and machine in pursuit of a goal defined by man Presumably, to be effective, this environment should allow the calculating speed, precision, and structured logical/iterative skill of the machine to serve the conceptual, intuitive, highly associative, and contexturally sensitive attributes of human mental function in the solution of problems Too often the match is obtuse, the goals vague, and the result frustration Much emphasis ought to be placed on the a priori specific problem oriented knowledge, gained through experience, which the man brings to the machine and wishes to share with it in a versatile but structured way

Suboptimal Sequential Decision Schemes With On-Line Feature Ordering

Abstract: Sequential decision algorithms with on-line feature ordering and a limited look-ahead approximation are considered. The algorithms can be used with or without contextual constraints for multicategory pattern recognition problems. Computational complexity due to on-line ordering of features is analyzed and related to system performance. Computer simulated results are obtained using a standard data set (Munson's multiauthor handprinted character files) and a careful test procedure.

The Pattern Recognition of Chinese Characters by Markov Chain Procedure

Abstract: The recognition and learning of Chinese characters are much more difficult than that of a check due to their complexity in the construction of features and lack of interrelationship and overall rules. The same words, written in different styles and habits by two persons, might be taken to be different words in the eyes of the scribes. It is important not only that the Chinese character be recognized, but also that it be recognized quickly and correctly by a simple program. An advanced procedure for recognition by use of the Markov chain procedure is presented.

Character pattern recognition method and apparatus using feature enhanced magnetic domain patterns

Abstract: A character pattern recognition method and apparatus for recognizing an unknown character pattern in accordance with a correlation between the unknown character pattern and each known standard character pattern, in which the unknown character pattern is temporarily stored in a magnetic thin plate in a form of a magnetic domain pattern. The magnetic domain pattern is modified by varying the bias magnetic field on the magnetic thin plate to provide a modified unknown character pattern emphasizing the feature of the unknown character pattern stored in the magnetic thin plate. The modified unknown character pattern is compared with each known modified feature enhanced standard character pattern to obtain the correlation.

Erratum: Optimisation of digital learning networks when applied to pattern recognition of mass spectra

Abstract: A pattern classifier employing n-tuple sampling digital learning networks is analysed to show that redundancy can occur both due to the common occurrence of sets of n-tuples of the sample pattern and invariant points in the patterns. Some experimental results are given for a mass-spectrum classifier, where the system has been optimised by reconnection to reduce this redundancy.

A Binary Feature Extraction Technique

Abstract: Numerous schemes are available for feature selection in a pattern recognition problem, but the feature extraction process is largely intuitive. A sequential feature extraction scheme is proposed for binary features. A decision function, which is linear and near optimal, is developed concurrently with each feature. Performance bounds are developed for several design strategies. Experimental results are given to illustrate the use of the scheme and the effectiveness of the performance bounds.

Performance Aspects of Stochastic Nonlinear System Classification by Pattern Recognition Methods

Abstract: The feasibility of identifying an unknown nonlinear stochastic system as belonging to a class of such systems by use of pattern recognition methods has recently been demonstrated. This paper examines the performance aspects of the classification technique by means of a theoretical approximation to the Bayes minimum classification error. The error is shown to be strongly dependent upon the pattern vector dimension thus showing that the classification error can be controlled by appropriate choice of pattern dimension. Two detailed classification examples are included, the first of which may be compared to earlier experimental results to substantiate the approximations used.

Computer Recognition of Facial Profiles

Abstract: : A system for the recognition of human faces from full profile silhouettes is described. The system is adaptively trained using a novel stack- oriented training procedure which is shown to be quite effective in identifying those feature vectors which are of most importance in the recognition process. Thus the training procedure generally produces authority files having a small number of entries. The feature vectors used are generated from a normalized autocorrelation function expressed in polar coordinates. These feature vectors are shown to be more effective in the recognition process than are the moment invariant features, at least for this problem. Experiments are described in which the system attains a recognition accuracy of 90% in a 10 class problem using 12-dimensional circular autocorrelation feature vectors. It is shown, by further experiments, that these results are no worse than a human observer's accuracy.

B74-5 Computer-Oriented Approaches to Pattern Recognition

Abstract: The book Computer-Oriented Approaches to Pattern Recognition by W. S. Meisel was recently reviewed very favorably by Prof. Breeding [1]. The review emphasized the wide range of numerical algorithms contained in the book, thus making it a desirable reference for anyone working in mathematical pattern recognition.