Showing papers on "Feature (machine learning) published in 1982"

Hidden Markov model speech recognition arrangement

Abstract: A speech recognizer includes a plurality of stored constrained hidden Markov model reference templates and a set of stored signals representative of prescribed acoustic features of the said plurality of reference patterns. The Markov model template includes a set of N state signals. The number of states is preselected to be independent of the reference pattern acoustic features and preferably substantially smaller than the number of acoustic feature frames of the reference patterns. An input utterance is analyzed to form a sequence of said prescribed feature signals representative of the utterance. The utterance representative prescribed feature signal sequence is combined with the N state constrained hidden Markov model template signals to form a signal representative of the probability of the utterance being each reference pattern. The input speech pattern is identified as one of the reference patterns responsive to the probability representative signals.

Cursive script recognition by elastic matching

Abstract: Dynamic programming has been found useful for performing nonlinear time warping for matching patterns in automatic speech recognition. Here, this technique is applied to the problem of recognizing cursive script. The parameters used in the matching are derived from time sequences of x-y coordinate data of words handwritten on an electronic tablet. Chosen for their properties of invariance with respect to size and translation of the writing, these parameters are found particularly suitable for the elastic matching technique. A salient feature of the recognition system is the establishment, in a training procedure, of prototypes by each writer using the system. In this manner, the system is tailored to the user. Processing is performed on a word-by-word basis after the writing is separated into words. Using prototypes for each letter, the matching procedure allows any letter to follow any letter and finds the letter sequence which best fits the unknown word. A major advantage of this procedure is that it combines letter segmentation and recognition in one operation by, in essence, evaluating recognition at all possible segmentations, thus avoiding the usual segmentation-then-recognition philosophy. Results on cursive writing are presented where the alphabet is restricted to the lower-case letters. Letter recognition accuracy is over 95 percent for each of three writers.

An algorithm for connected word recognition

Abstract: The principles of an efficient one-pass dynamic programming whole-word pattern matching algorithm for the recognition of spoken sequences of connected words are described. Particular attention is given to the technique for keeping track of word-sequence decisions, which may be constrained by a finite-state syntax. Some extensions of the technique are discussed.

Computational Geometric Problems in Pattern Recognition

Abstract: This paper surveys recent results in the design and analysis of algorithms for solving geometric problems in pattern recognition. Among the problems considered are: the convex hull, the diameter, Voronoi diagrams, the relative neighborhood graph, polygon decomposition, and distance between sets. Some new results are presented; among them a new 0(n) algorithm for merging two convex polygons and a proof that a convex hull algorithm of Kim and Rosenfeld (35) works. Several open problems are also mentioned.

31 Pattern recognition in chemistry

Abstract: Publisher Summary Multivariate methods of pattern recognition, classification, discriminant analysis, factor and principal components analysis and the like have been found most useful in many types of chemical problems. With increasing experience of multivariate data analysis, chemists have reached a better understanding of the nature of chemical data and, thereby, have been able to specify more explicitly which kind of information one wants to extract from the data. This chapter describes this understanding of how to apply pattern recognition in chemistry. The multivariate nature of chemical measurements generated by modern chemical instrumentation together with the nature of chemical theory which involves unobservable “micro”-properties make a strong case for a rapidly increased use of multivariate data analysis including various methods of pattern recognition in chemistry. With the availability of fast, inexpensive and graphics-oriented computers, the large number of calculations is no longer a problem.

Method and apparatus for speech recognition and reproduction

Abstract: Speech signal analysis for data reduction, as stored for synthesis or recognition, is improved by features including: digital spectral analysis; reduction of channel data and bit allocation by selective summation of groups of contiguous data; using the mean average of the log amplitude to find the deviation for each channel; also using the instaneous shape of the mean value for each channel for pairs of adjacent frames, all combined to find a feature ensemble for each pair of adjacent frames.

Development of an automatic identification system of spoken languages: Phase I

Abstract: The feasibility of a new approach to automatic language identification is examined in this pilot study The procedure involves the application of pattern analysis techniques to features extracted from the speech signal The database of the extracted features for five speakers from each of eight languges was divided into a learning subset and an evaluation subset A potential function was then generated for all features in the learning subset The complexity of the decision function was systematically increased until all members within the learning subset could be separated into the properly identified languages Although the constraints on this pilot study necessarily precluded feature ordering and selection, the application of the decision function to the evaluation subset resulted in an over-all 84% classification accuracy

Speech recognition system and method

Abstract: Speech recognition accuracy is significantly enhanced by employing recognition criteria that involve comparison (in blocks 74, 84), as between spoken command words and stored "training" words, of both time-dependent feature arrays (72,73) and time-independent feature vectors (82, 83). The novel time-independent feature vector for any given word consists of a representation of the frequency of occurrence of each particular feature at any of several "time slots" in the word. The extra information about the word gained by a comparison (84) of its vector with a corresponding vector of each training word assists in the final decision (90).

Extraction of information from large data sets by pattern recognition

Abstract: Pattern recognition permits to extract information present in large data sets in an automatic way. Many scientists acknowledge this fact but are rebutted by the task of learning to use pattern recognition methods. Indeed, there are many methods available and for the newcomer it is extremely difficult to make a selection. For this reason, the paper starts by explaining the models used in pattern recognition. This is followed by a critical discussion of advantages and disadvantages of the methods and a selection of preferred methods.

Principal-Component Imagery For Statistical Pattern Recognition Correlators

Abstract: Concepts, measures, and models of image quality are shown to be quite important in pattern recognition applications. Pattern recognition of imagery subjected to geometrical differences (such as scale and rotational changes) and intensity differences (such as arise in multispectral imagery) are considered. After modeling these image differences as a stochastic process, the optimal filter is derived. This filter is shown to be the principal component of the data. This pattern recognition algorithm is verified using multi-sensor imagery, and the results are found to compare favorably to those obtained using other candidate techniques.

A connectionist approach to visual recognition

Abstract: Strictly sequential approaches to computer vision are at best slow and cumbersome, at worst impossible. In this thesis, a massively parallel, connectionist approach is brought to bear on the problem of visual recognition. Computing with connections is a synthesis of results from Neuroscience, Computer Science, and Psychology. The fundamental assumption of connectionism is that individual computing units do not transmit large amounts of symbolic information. Instead, these units compute by being appropriately connected in a network of similar units. Using the communication pathways (connections) defined by the arcs of the network, the units cooperate and compete towards a globally consistent interpretation of the input scene. The problem, visual recognition, is defined as matching instances of predefined objects in the input with a fixed set of internal models. Predefined objects come from Kanade's Origami World{Kanade78}. The program represents and recognizes such pre-defined objects from line drawing input. To organize these networks, conceptual hierarchies are defined. A conceptual hierarchy is a semantic network hierarchically arranged according to abstraction levels. Levels represent the extraction of progressively more complex features. A node on a level, a computing unit, represents an instantiation of a feature defined on that level. Connections represent the composition and competition relations between feature units. Iterative relaxation is the form of control in the network. Each unit iteratively computes activation levels, a reflection of current confidence in the associated feature. Numerous testcases illustrate network behavior in presence of perfect, noised, incomplete and occluded input. The greatest benefits of this approach are seen in the systems ability to cope with incomplete and occluded input. Another advantage is an inherently parallel approach that eliminates the need for establishing a search order on exploring interpretations. The system exhibits many similarities with the structure and behavior of animal vision systems.

Fusion of Multisensor Data in Pattern Recognition

Abstract: This paper describes a number of techniques by which to fuse multisensor data (images, signals, scenes, etc) and by which to generate higher level representations of an unknown pattern within the context of pattern recognition The basic steps involved are: A locate a representation based on exogenous context information B compare two representations to find out if they refer to the same entity C merging features from two representations of the same pattern into a new feature D aggregaring two representations into a higher level representation

Pattern Recognition for Automatic Visual Inspection

Abstract: Three major approaches to pattern recognition, (1) template matching, (2) decision-theoretic approach, and (3) structural and syntactic approach, are briefly introduced. The application of these approaches to automatic visual inspection of manufactured products are then reviewed. A more general method for automatic visual inspection of IC chips is then proposed. Several practical examples are included for illustration.© (1982) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

Recognition of speech or speech-like sounds

Abstract: An improved method of speech recognition suitable for use in simple type speech recognition systems is disclosed. The method of speech recognition uses short time self-correlation functions as feature parameters for recognition of speech or speech-like words and especially effecting preliminary selection utilizing part of data for final recognition, that is, short time self-correlation functions of lower degrees (typically, primary to cubic). In carrying out recognition of speech or speech-like sounds, the method involves creating self-correlation functions for input sound signals, deciding the intervals of the sound signals, normalizing the time axies in conjunction with the sound intervals, and conducting recognization of words or the like through deciding using the self-correlation functions as feature parameters whether there is matching with reference patterns. In the above method, preliminary selection is effected prior to the final recognization step by means of linear matching using the self-correlation functions of lower degrees.

Method of speech recognition

Abstract: In a speech recognition system, similarity calculations between speech feature patterns are reduced by stopping similarity calculations for any one reference pattern when a frame in the pattern fails to exceed a corresponding similarity threshold.

A Syntactic Approach to Seismic Pattern Recognition

Abstract: The nearest-neighbor decision rule for syntactic patterns is applied to seismic pattern classification. Each pattern is represented by a string. The string-to-string distance is used as a similarity measure. Another method using finite-state grammars inferred from the training samples and error-correcting parsers is also implemented. Both methods show equal recognition accuracy; however, the nearest-neighbor rule is much faster in computation speed. The classification results of real earthquake/explosion data are presented.

Theory and Model of the Human Global Analysis of Visual Structure

Abstract: An introduction is given to a theory of global structure analysis in early visual information processing. The theory relates retinal, visual cortical, retino-collicular, cortico-collicular, and oculomotor mechanisms to specific properties of the (human) visual structure analysis. Information theoretic modeling of the retina, particularly rate distortion theoretic interpretations of the retina-oculomotor system functions reveal both behavior and information processing task optimality. From a structure system point of view these properties bring forth a global structure identification function and its supporting global data systems which extract information from the object system. A general formulation of the structure identification function and its data systems is given in terms of partial correlations of perceptual similarity. The theory has been implemented for dot patterns through a set of computer programs. This computer model determines structure?perceptually important global features, their "strength," and the pattern regions that give rise to these features. Psychophysical measurements were necessary to set the "human" parameters of the model. Global analysis of visual structure is considered to be essentially a structure segmentation and feature classification task that precedes detailed information extraction. Theory-based interpretation of the functional organization of the sensory, motor, and neurophysiological components of the visual system leads to expectations concerning a scan strategy that directs the retina towards regions of high structure information density.

Automatic recognition of syllabic speech segments using spectral and temporal features

Abstract: An automatic speech recognition system is presented which starts from a demisyllable segmentation of the speech signal. Recognition of these segments is based on a set of spectral and temporal acoustic features which are automatically extracted from LPC-spectra and assembled in one feature vector for each demisyllable. The 24 components of this vector describe formants, formant loci, formant transitions, formant-like "links" for characterization of nasals, liquids or glides, the spectral distribution of fricative noise or bursts (turbulences), and duration of pauses. Preliminary recognition experiments were carried out with feature vectors extracted from a set of 360 German initial demisyllables which represent 45 consonant clusters combined with 8 vowels. When compared with template matching methods, the feature representations yield a drastic reduction in the number of components needed to represent each segment.

Character recognition apparatus

Abstract: The invention provides a character recognition apparatus having an original feature extracting section for extracting as an original feature a feature deliberately neglected in the processes of pre-process conversion and recognition feature extraction. The original feature extracted by the original feature extracting section is used for final recognition of a set of characters, thus preventing erroneous character recognition.

Geometric character recognition using skeleton and stroke width representation

Abstract: character recognition system is shown which makes use of skeleton and stroke width data in digital form. Data representing certain features such as curves, lines and loops of the skeleton are extracted by a feature extractor to provide character codes representing the digits "0" through "9". The character codes are compared then with reference standards in a look-up table which provides a classification of the character codes according to whether there is (1) a direct match with standard character codes in the look-up table, (2) a "confused class" indicating that the particular character code represents more than one character or class of data in the table or (3) a "noise code" indicating the particular character code cannot be identified from data in the table. In case of (1) above the identification is transmitted for immediate use. In cases (2) and (3) further analysis of characteristics and modifications of characteristics of the lines, curves and loops as well as analysis of stroke width are employed to enable identification of additional character codes.

Speaker recognition using a feature weighting technique

Abstract: This paper describes a technique for increasing the ability of a text-dependent speaker recognition system to discriminate between speaker classes; this technique is to be performed in conjunction with the nonlinear time alignment between a reference pattern and a test pattern. Unlike the standard approach, where the training of the recognition system merely consists of storing and averaging or selecting the time normalized training patterns separately for each class, the training phase of the system is extended in that a weight is determined for each individual feature component of the complete reference pattern according to the ability of the feature to distinguish between speaker classes. The weights depend on the time axis as well as on the frequency axis. The overall distance computed after nonlinear time alignment between a reference pattern and a test pattern thus becomes a function of the given set of weights of the reference class considered. For each class, the optimum weights result from the ideal criterion of minimum error rate. Instead of this criterion, the closely related but mathematically more convenient Fisher criterion is used that leads to a closed from solution for the unknown weights. Based on these weights, the selection of subsets of effective features is studied in order to further improve the class discrimination. The feature weighting and selecting techniques are tested using a data base of utterances recorded off dialed-up telephone lines. The experiments indicate that feature weighting and feature selection can reduce the error rates by a factor of two or more both for speaker identification and speaker verification.

Towards Very Large Vocabulary Word Recognition

Abstract: : In this paper, preliminary considerations and some experimental results are presented in an effort to design Very Large Vocabulary Recognition (VLVR) systems. We will first consider the applicability of current recognition techniques and argue their inadequacy for VLVR. Possible alternate strategies will be explored and their potential usefulness statistically evaluated. Our results indicate that suprasegmental cues such as syllabification, stress patterns, rhythmic patterns, rhythmic patterns and the voiced - unvoiced patterns in the syllables of a word provide powerful mechanisms for search space reduction. Suprasegmental feature could thus operate in a complementary fashion to segmental features.

Statistical Pattern Recognition of Binary Variables

Abstract: Many different methods for tackling binary variable pattern recognition problems have been proposed. The differences arise from the particular special needs of the individual problems as well as the background and orientation of the authors presenting the methods. This paper summarises these methods, beginning with the basic multinomial method and showing how other methods can be seen as different ways of tackling the shortcomings of the basic method by imposing different types of dependence structure on the data. The various merits and demerits of the methods are pointed out in the hope that this will aid in choice of method.

16 Structural methods in image analysis and recognition

Abstract: Publisher Summary A wide variety of techniques has been developed for image analysis. One common approach consists of image segmentation using some type of similarity criterion for grouping areas within an image followed by measurement of resulting region properties such as shape and texture. These measurements are used to classify the regions into types by computing the similarity of these measurements to those of a set of tracking regions. Statistical methods such as discriminant analysis and the Bayesian classifiers have been used for this classification step. This chapter focuses on structural methods in image analysis and recognition. The primary goal of structural pattern recognition procedures has been the recognition of objects in an image. Structural methods are appealing because they allow the designer or the user of a pattern recognition system to employ a somewhat intuitive description of an object as the basis for a recognition scheme. A fundamental part of many structural recognition systems is a search space. This space may be explicitly stored in a computer or implicitly stored and dynamically generated as in a grammar. Often measures of merit are defined on those parts of the search space that have been examined.