Showing papers on "Feature (machine learning) published in 1979"

Two-level DP-matching--A dynamic programming-based pattern matching algorithm for connected word recognition

Abstract: This paper reports a pattern matching approach to connected word recognition. First, a general principle of connected word recognition is given based on pattern matching between unknown continuous speech and artificially synthesized connected reference patterns. Time-normalization capability is allowed by use of dynamic programming-based time-warping technique (DP-matching). Then, it is shown that the matching process is efficiently carried out by breaking it down into two steps. The derived algorithm is extensively subjected to recognition experiments. It is shown in a talker-adapted recognition experiment that digit data (one to four digits) connectedly spoken by five persons are recognized with as high as 99.6 percent accuracy. Computation time and memory requirement are both proved to be within reasonable limits.

Multiple template speech recognition system

Abstract: A speech analyzer for recognizing an unknown utterance as one of a set of reference words is adapted to generate a feature signal set for each utterance of every reference word. At least one template signal is produced for each reference word which template signal is representative of a group of feature signal sets. Responsive to a feature signal set formed from the unknown utterance and each reference word template signal, a signal representative of the similarity between the unknown utterance and the template signal is generated. A plurality of similarity signals for each reference word is selected and a signal corresponding to the average of said selected similarity signals is formed. The average similarity signals are compared to identify the unknown utterance as the most similar reference word. Features of the invention include: template formation by successive clustering involving partitioning feature signal sets into groups of predetermined similarity by centerpoint clustering, and recognition by comparing the average of selected similarity measures of a time-warped unknown feature signal set with the cluster-derived reference templates for each vocabulary word.

Multidimensional letter similarity derived from recognition errors.

Abstract: In order to provide a reliable measure of the similarity of uppercase English letters, a confusion matrix based on 1,200 presentations of each letter was established. To facilitate an analysis of the perceived structural characteristics, the confusion matrix was decomposed according to Luce’s choice model into a symmetrical similarity matrix and a response bias vector. The underlying structure of the similarity matrix was assessed with both a hierarchical clustering and a multidimensional scaling procedure. This data is offered to investigators of visual information processing as a valuable tool for controlling not only the overall similarity of the letters in a study, but also their similarity on individual feature dimensions.

Problem-Solving Models and Search Strategies for Pattern Recognition

Abstract: Noting the major limitations of multivariate statistical classification and syntactic pattern recognition models, this paper presents an overview of some recent work using alternate representations for multistage and nearest neighbor multiclass classification, and for structural analysis and feature extraction. These alternate representations are based on generalizations of state-space and AND/OR graph models and search strategies developed in artificial intelligence (AI). The paper also briefly touches on other current interactions and differences between artificial intelligence and pattern recognition.

On-line pattern recognition system for hand-written characters

Abstract: A pattern recognition system operating on an on-line basis for hand-written characters, in particular for hand-written Chinese characters comprises a character input unit for providing the coordinates of a plurality of points on the strokes of a written input character, a classification unit for classifying the input characters to the first group having equal to or less than three strokes, and the second group having equal to or more than four strokes, an approximate unit for providing a plurality of feature points to each of strokes, the number of strokes being six for each stroke in the first group of characters and three for each stroke in the second group of characters, a pattern difference calculator for providing the sum of the length between the feature points of the input character and those of the reference characters which are stored in the reference pattern storage, and a minimum difference detector for determining the minimum length among the pattern differences thus calculated. The input character is recognized to be the same as the reference character which provides said minimum length.

Adaptation Algorithms for Binary Tree Networks

Abstract: The automatic synthesis of Boolean switching functions by adaptive tree networks is discussed. The concept of heuristic responsibility, by means of which parts of a tree become specialized to certain subsets of input vectors, is explained. Applications to pattern recognition and optical character recognition (OCR) problems are described.

Spectral classification of phonemes by learning subspaces

Abstract: A new pattern recognition scheme with learning ability is introduced, and its application to the labeling of phonemes is reported. The basic classification algorithm is known as the subspace method in which classes of patterns are defined as linear vector subspaces spanned by the prototypes, and the class affiliation of an unknown pattern vector is decided by comparison of its orthogonal projections on the various subspaces. This method is here modified in two ways. In one of them, the prototype patterns are selected conditionally according to classification results obtained during training. In the second modification the subspaces are rotated in proper directions in the training procedure, depending on the classification results. By means of these methods, for the average accuracy of classification with 15 phonemic classes from continuous Finnish speech, a value of about 80 per cent was obtained.

Pattern recognition with a warping function decided for each reference pattern by the use of feature vector components of a few channels

Abstract: In a pattern recognition device according to pattern matching, one or more specific dimensions of vector components are memorized for each reference pattern feature vector sequence in a reference pattern memory for the reference pattern feature vector sequences. A warping function for time-normalizing input pattern feature vectors of a sequence and the vectors of each reference pattern feature vector sequence is determined so as to minimize the difference between a pattern represented by the specific vector components of the specific dimension or dimensions and another pattern represented by the vector components corresponding in the input pattern feature vector sequence to the specific reference pattern feature vector components as regards the dimensions of a space in which each input or reference pattern feature vector is defined. The input pattern feature vector sequence and each reference pattern feature vector sequence are subjected to nonlinear pattern matching with reference to the warping function. The pattern matching may be between the vector components of all dimensions or those of several dimensions including the specific dimension or dimensions. Preferably, one or more dimensions are specified as the specific one or ones by selecting each dimension for which a variation with time of a pattern represented by the reference pattern feature vector components is a maximum of similar variations of patterns represented by the vector components of other dimensions.

A Syntactic-Semantic Approach to Image Understanding and Creation

Abstract: A syntactic-semantic approach to information extraction from images is described. The methodology involves the injection of semantic considerations into context-free grammars. The semantic considerations include feature vectors, selection restrictions, feature transfer functions, semantic well-formedness, etc. With such injection, we can make a description scheme which carries the numerical, the structural, and the a prior real world knowledge about the pattern we want to extract. From the description we can construct an analytical mechanism, the creation machine, which wil find the desired pattern amid a chaos of noisy primitives.

Inference of a knowledge source for the recognition of nasals in continuous speech

Abstract: A system for the automatic recognition of bilabial /m/ and alveolar /n/ in vowel-consonant-vowel utterances extracted from continuous speech is presented. It is based on a syntactic pattern recognition approach and the use of fuzzy relations for evaluating phonemic hypotheses. The knowledge source, based on very simple transition networks with associated simple semantic rules, is inferred from experiments. Results obtained for four male speakers are presented together with an acoustic-phonetic motivation of the approach used. These show how accounting for coarticulation effects gives substantially better performances than previous approaches.

A Pattern Deformational Model and Bayes Error-Correcting Recognition System

Abstract: A pattern deformational model is proposed in this paper. Pattern deformations are categorized into two types: local deformation and structural deformation. A structure-preserving local deformation can be decomposed into a syntactic deformation followed by a semantic deformation, the former being induced on primitive structures and the latter on primitive properties. Bayes error-correcting parsing algorithms are proposed accordingly which not only can perform normal syntax analysis but also can make statistical decisions. An optimum Bayes error-correcting recognition system is then formulated for pattern classification. The system can be considered as a hybrid pattern classifier which uses both syntactic and statistical pattern recognition techniques.

Automatic Classification of Spirometric Data

Abstract: Pattern recognition principles have been applied to 200 sets of spirometric data obtained from pulmonary function laboratory patients. Each patient was classified by a pulmonary specialist as normal, restricted, or mildly, moderately, severely, or very severely obstructed. Each patient was represented by a five-element pattern vector consisting of forced vital capacity (FVC), forced expiratory volume in one second (FEV1), midmaximum flow rate (MMFR), and flow rates with 50 and 25 percent of the vital capacity remaining (V?50 and V?25) normalized by predicted values. By Karhunen-Loeve expansion techniques, this vector was reduced to a two-feature pattern vector with only a 6 percent residual mean square representation error. The more important feature essentially represented the average of the three flow rates, while the second feature depended on FVC and FEV1. Data were divided into training and testing sets, and using the former, a parametric Bayes classifier and one-and two-layer pair-wise Fisher linear classifiers, were designed to assign patterns described by the two derived features to one of the six categories. With the testing set, overall recognition rates were 81 to 82 percent, with most errors representing misclassifications within the four obstructive categories. If the four obstructive classes were considered as a single class, the recognition rate increased to about 94 percent.

Linguistic analysis of experimental curves

Abstract: In the linguistic approach to pattern recognition, a special-purpose language is constructed, and algorithms using this language to analyze objects are developed. The algorithms described in this paper are for the analysis of experimental curves. A function to estimate the "complexity" of the curve parts is proposed. The algorithms differ from those for image processing in several respects. A significant feature of the computer-generated language is the good interpretability of its words by humans.

The global qualitative study of solutions to a conservation law (ii)

Abstract: In this paper, the classes of initial value functions broader than those in [1] are considered. The geometric feature of the shock waves which stand as a row of "trees" spreading to the infinity, is clearly exhibited. The asymptotes in various cases are formulated. Particularly the C~∞ initial value functions with compact suppcrts are considered and an exact estimate about the support lengths of the solutions is obtained. C~∞ periodic initial value functions with arbitrary means are investigated, and an exact estimate for the decay of solutions is done. Further, the generic properties of any classes of initial data which appear in this paper are also studied.

Speech recognition system

Abstract: PURPOSE:To increase greatly a processing speed by collation-selecting a recognition- expectant word from the reference contraction feature quantity previously registered on the basis of contraction feature quantities obtained in respective sampling processes, and by collating the selected word with an unknown voice. CONSTITUTION:Syllable representative-point extraction circuit part 6 counts up the number of accumulation during the determination of uneven sampling points T0, T1..., and determines a fixed number of sampling points from the maximum-weight uneven sampling point in the order regarding discrete values of respective uneven sampling points T0, T1... as weights. Namely, a sampling point which corresponds to a syllable is determined and on the basis of the contraction feature quantity obtained corresponding to each sampling point, recognition expectant word is selected 10 by collation from the previously-registered reference contraction feature quantity. Then, an unknown input voice is collated with the selected recognition expectant word. Consequently, the processing speed can be increased greatly.

Local versus global decisions in image recognition

Abstract: This paper discusses the ways of finding consistency between the well-known statistical statement that "guessing destroys information" and the practically obvious advantage of hierarchical decisions. Certain nonstatistical sources of recognition errors are indicated, the influence of these sources increasing with the size of the image parts on which the first-stage discrete decisions are taken. The rejection criterion is examined from the statistical point of view and the necessity of mathematical models for all images to be rejected is demonstrated. The analysis of the possibilities for developing models of both images to be recognized and to be rejected leads to the conclusion that image recognition should be realized by hierarchical systems. An example of a working hierarchical recognition system for interpretation of handmade drawings is described.

Appropriateness of different rule types in speech synthesis

Abstract: The Cornell Speech Research System has been augmented to support a third kind of rule. In addition to feature and parameter rules, the user can now express letter‐to‐sound rules. Feature rules modify the feature composition of an utterance, and parameter rules transform the output of the feature rules into a file of synthesizer parameters. Letter‐to‐sound rules convert standard orthography into a phonetic transcription. A powerful notation has been developed for expressing these rules. By using the system, we have found that the three kinds of rules can be written to interact in linguistically meaningful ways to yield speech.

Structure and performance of an on-line speaker verification system

Abstract: During the last years an on-line system for automatic speaker verification has been developed at the Heinrich Hertz Institute. The structure of this system is described and the techniques used for preprocessing, feature extraction and classification are discussed in detail. Results are based on speech material gathered from "true speakers" and "impostors" who tried to imitate the true speakers supported by internal system information. It is shown that it is necessary to combine several feature sets to obtain a high system performance.

Coin-operated gaming machine - has partly differently designed game feature carriers moving freely in observable playing field for recognition

Abstract: The coin-operated gearing machine has symbols marked on its game feature carries According to the final position of these symbols in a recognition zone wins can be given to the player The game feature carriers (9) can be whirled within the playing field (4, 5, 6) by means of air The ballon-shaped playing field emerges with its neck in the recognition zone, at the bottom of which a nozzle is provided for an air supply This nozzle may be connected to a compressor via a solenoid valve This valve may be connected to a control switch

Theory of self-organizing nerve nets with special reference to association and concept formation

Abstract: The present paper proposes a mathematical theory of self-organizing nerve nets, which is applicable to various types of supervised and unsupervised learning, such as learning decision, concept formation, association, etc. Given an environmental information source, a neural system automatically forms a number of separate routines to process the signals in it. This kind of unsupervised self-organization underlies commonly in formation of categories, feature extractors, and content addressable memories. This problem is analyzed mathematically, as well as models of topographic organization of nerve fields and of associative memories, by the proposed method.

Multistage decision schemes for speaker recognition

Abstract: Speaker recognition schemes which work satisfactorily for small populations often fail when the number of classes is very large One way of solving such problems is to go in for multistage classification schemes The basic technique is to successively reduce the number of classes in several stages using one feature at each stage and when the number of classes is less than a predetermined value then the final decision is made The whole scheme is designed so that the probability of error is fixed at an acceptable level The computational cost of such a multistage scheme depends on the features used at each stage and the cost of measurement of each feature The features to be used at each stage are determined so as to reduce the average computational cost for making a decision This procedure is formulated as a stochastic optimal control problem and is illustrated by designing speaker recognition system for 60 speakers The overall accuracy of the system is 972 %

Recognition and processing system of character pattern

Abstract: PURPOSE:To prevent recognition from being disabled and to improve recognition precision by performing postprocessing by using language knowledge after recognition processing based upon the processed features of a character pattern, and collating the last and following a character pattern to perform the recognition processing when the unkown pattern does not coincide with only one code. CONSTITUTION:An unknown input character pattern string 1 is outputted from a photoelectric transducer 2 as an electric signal proportional to the pattern and compared with a reference value by a binary-coding circuit 3 to be converted into a black-and- white binary signal, which is applied to a classifying circuit 4. The circuit 4 compares this with a standard pattern to select the kind of the character and extract the feature amount and the distance value of each pattern, and they are stored in a storage circuit 5. A language processing circuit 6 collates the last and following recognition results among candidate character kinds stored in the circuit 5 with the linguistic knowledge stored in a storage circuit 7. Then, a collating circuit 8 a candidate character which is collated with the linguistic knowledge by the circuit 6 with a candidate character kind from the circuit 5, and when the unknown pattern does not coincide with only one code, the last and following character patterns are collated to improve the recognition precision.

Formal objects and feature associations in ARGOS-II

Abstract: ARGOS-II is a problem solving system developed at L.S.I, laboratory. Its purpose is to embody a general model of the decision function of a robot. After a brief global characterization of the system, an overview of the initial specialization process is provided. Then two interesting particularities of the system are emphasized : "formal prospective aptitude" and "automatic control of feature associations". ARGOS-II generates action plans but also performs the execution monitoring. ARGOS-II is a pattern directed inference system. The "formal prospective aptitude" allows to continue to build plans even when some circumstances are unknown. They permit to avoid some awkwardnesses or failures and are open to learning. Not only does the "feature associations" mechanism deal with the heavy problem of updating the attributes of a group of objects involved in a unique action, but it also allows to easily create powerful demon functions useful for Artificial Intelligence programming.

Statistical image processing and recognition

Abstract: This paper is concerned with the methodologies in statistical image processing and recognition. Specific areas considered are the following: (1) The decision rules in image recognition and their comparative evaluation under finite sample size condition; (2) Statistical feature extraction techniques for image segmentation with emphasis on the statistical characteristic of textural features; (3) Statistical contextual analysis algorithms for images. Emphasis is placed on the contextual pre processing/postprocessing techniques to implement the optimum decision rules with context; (4) Statistical image modelling techniques including the nonhomogeneous models and the autoregressive models. The software problems involved in these areas are also examined in details.

Two-Level DP- Programming-Based Patter ri for Connected

Abstract: This paper reports a pattern matching approach to con- nected word recognition. First, a general principle of connected word recognition is given based on pattern matching between unknown continuous speech and artificially synthesized connected reference patterns. Time-normalization capability is allowed by use of dynamic programming-based time-warping technique @P-matching). Then, it is shown that the matching process is efficiently carried out by breaking it down into two steps. The derived algorithm is extensively subjected to recognition experiments. It is shown in a talker-adapted recognition experiment that digit data (one to four digits) connectedly spoken by five persons are recognized with as high as 99.6 percent accuracy. Computation time and memory requirement are both proved to be within reasonable limits. is easily performed by updating the reference pattern. Further- more the pattern matching method works fairly well for any foreign language. In this paper pattern matching is carried out between speech patterns with quasi-continuous time axes. Here, the term "quasi-continuous'' means that time sampling is made with a relatively short (10 ms or so) constant period. It is well known that the resulting speech pattern is then very redundant. From the computational economy stand- point, quasi-continuous sampling will be insufficient. Some phonemic level segmention is naturally expected to reduce the redundancy. In spite of these well-known factors, preliminary segmentation is utterly discarded in this investigation. It is considered that computational efficiency improvement is a secondary problem, coming after the establishment of a more reliable recognition principle. Two-level DP-matching discussed in this paper is a connected word recognition method based on pattern matching. It employs no preliminary segmentation in principle. Pattern matching is made on a whole connected word basis, that is, between unknown continuous and artificially synthesized connected reference patterns. The time-normalizing feature is given through the use of dynamic programming. It is shown that the matching process is efficiently carried out by breaking it down into two steps, word level matching and phrase level matching. Two execution algorithms are derived which are suitably applicable to the computer simulation experiment and the real-time recognition system, respectively. High-level recognition performance was established through several connected word recognition experiments. Computation time and memory requirements are both estimated to be within reasonable limits.

Identification of Parabolic Distributed-Parameter Systems by a Pattern-Recognition Approach

Abstract: A pattern-recognition method is given for identification of distributed parameter systems This approach takes care of the system characterization which is the first and critical step in the identification problem Systems which can be represented by parabolic partial differential equations are considered The pattern recognition system has been designed to classify the pattern measurements into identifiable classes The inputs to this pattern recognition system are the data concerning the inputs and observation responses of the distributed systems and the output is the classification of the distributed parameter system