Showing papers on "Feature (machine learning) published in 1986"

Fuzzy Mathematical Approach to Pattern Recognition

Abstract: This book aims to present results of investigations, both experimental and theoretical, into the effectiveness of fuzzy algorithms as classification tools in some problems concerned with the field of pattern recognition and image processing. Compares results to those obtained with statistical classification techniques.

A Step Towards Unification of Syntactic and Statistical Pattern Recognition

Abstract: The problem of pattern recognition is discussed in terms of single-entity representation versus multiple-entity representation. A combined syntactic-semantic approach based on attributed grammars is suggested. Syntax-semantics tradeoff in pattern representation is demonstrated. This approach is intended to be an initial step toward unification of syntactic and statistical approaches to pattern recognition.

Exploring recognition with interchanged facial features

Abstract: Any attempt to unravel the mechanism underlying the process of human face recognition must begin with experiments that explore human sensitivity to differences between a perceived image and an original memory trace. A set of three consecutive experiments are reported that were collectively designed to measure the relative importance of different facial features. The method involved the use of image-processing equipment to interchange cardinal features among frontally viewed target faces. Observers were required to indicate which of the original target faces most resembled the modified faces. The results clearly establish the dominant influence of the head outline as the major recognition feature. Next in importance is the eye/eyebrow combination, followed by the mouth, and then the nose. As a recognition feature in a frontally presented face, the nose is hardly noticed. The number of apparently random responses to some faces indicates that a surprisingly different face can sometimes arise from a fortuitou...

Memory storage and retrieval processes in category learning.

Abstract: The detailed course of learning is studied for categorization tasks defined by independent or contingent probability distributions over the features of category exemplars. College-age subjects viewed sequences of bar charts that simulated symptom patterns and responded to each chart with a recognition and a categorization judgment. Fuzzy, probabilistically defined categories were learned relatively rapidly when individual features were correlated with category assignment, more slowly when only patterns carried category information. Limits of performance were suboptimal, evidently because of capacity limitations on judgmental processes as well as limitations on memory. Categorization proved systematically related to feature and exemplar probabilities, under different circumstances, and to similarity among exemplars of categories. Unique retrieval cues for exemplar patterns facilitated recognition but entered into categorization only at retention intervals within the range of short-term memory. The findings are interpreted within the framework of a general array model that yields both exemplar-similarity and feature-frequency models as special cases and provides quantitative accounts of the course of learning in each of the categorization tasks studied.

Pattern recognition method and apparatus

Abstract: A general purpose pattern recognition method and apparatus comprises a hierarchical network of basic recognizers, each basic recognizer being capable of discriminating a particular feature at a lower level and providing outputs for higher levels of abstraction. In a learning mode, a series of sample patterns having a feature are presented as input along with several near-miss patterns. The pattern recognition apparatus learns to recognize the feature by keeping track of which basic recognizers detect patterns containing the feature. In a recognition mode, the invention determines if a presented pattern has the feature by polling the basic recognizers. A summation algorithm calculates the likelihood that the presented pattern has a particular feature.

Speech recognition for recognizing the catagory of an input speech pattern

Abstract: Speech recognition method and system are adapted to previously prepare a noise pattern in response to environmental noise prior to inputting a speech signal, evaluate a speech feature vector Bi yielded by subtracting the noise pattern from a feature vector Ai of the input speech upon inputting the speech signal thereafter, spectrum-normalize the speech feature vector Bi, evaluate a local peaks vector by making use of binary-coding processing wherein only a component of a channel being the maximum of the spectrum-normalized vector in a direction of frequency is assumed to be "1", evaluate pattern similarity between an input pattern comprising a local peaks vector from the start point to the end point of the input speech and previously prepared reference patterns of a plurality of categories of the same format as the input pattern, and judge a category of a reference pattern being the maximum among the pattern similaritites as a recognized result. As a result, any influence of environmental noise on the speech in concern is eliminated for assuring highly accurate recognition. Furthermore, the speech recognition method and system are adapted to improve efficiency and accuracy of speech recognition by making use of processing of evaluating a new speech feature vector of a prescribed frame number of effecting linear expansion therefor on the basis of internal division of adjoining speech feature vectors Bi and Bi+1, and processing of smoothing a local peaks vector.

On hidden Markov models in isolated word recognition

Abstract: Hidden Markov modeling has become an increasingly popular technique in automatic speech recognition. Recently, attention has been focused on the application of these models to talker-independent, isolated-word recognition. Initial results using models with discrete output densities for isolated-digit recognition were later improved using models based on continuous output densities. In a series of experiments on isolated-word recognition, we applied hidden Markov models with multivariate Gaussian output densities to the problem. Speech data was represented by feature vectors consisting of eight log area ratios and the log LPC error. A weak measure of vocal-tract dynamics was included in the observations by appending to the feature vector observed at time t, the vector observed at time t-δ, for some fixed offset δ. The best models were obtained with offsets of 75 or 90 msecs. When a comparison is made on a common data base, the resulting error rate of 0.2% for isolated-digit recognition improves on previous algorithms.

Character recognition method and system capable of recognizing slant characters

Abstract: A character recognition system includes a scanner for scanning a character to be processed to produce a character data, which is processed in a predetermined manner to extract a character feature, preferably contour information of the character Preferably, a plurality of predetermined directionality codes each associated with a unique pixel arrangement pattern are prepared and they are assigned to the contour depending on the local pixel arrangement condition The contour information defined by such directionality codes is further processed to define a histogram or feature vector of the directionality codes which is compared with a first group of reference characters without inclination and a second group of reference characters with inclination When a predetermined condition is met, then the mode of operation is switched such that an unknown character is compared with only one of the first and second groups of reference characters

Discrete Attribute Models of Brand Switching

Abstract: A formal analysis is presented for three general classes of discrete attribute models of brand switching. The analysis focuses on the role of feature importance in specifying transition probabilities. A number of formal properties based on ordinal relations between transition probabilities are defined and each class of models is shown to satisfy a unique subset of the properties. The analytic results reveal important relations between the functional forms of model equations and the managerially-oriented interpretations that are given to the variables in those equations.

The Use of Orthographic Knowledge in Beginning Reading

Abstract: The purpose of the present study was to determine the relationship between word recognition ability, knowledge of orthographic structures, and use of orthographic knowledge in word recognition. Fifty-six first and second graders were administered a word recognition test, two tests of orthographic knowledge, and two search tasks. The results indicated that when searching for multiple word targets children with word recognition levels of less than 2-2 searched similarly through all fields, whereas children with word recognition levels of 2-2+ searched faster through pseudowords and nonwords than through words. When searching for members of a category, children with word recognition levels below 2-1 searched faster through nonwords and pseudowords than through words providing no evidence for the use of orthography in word search. Children with word recognition levels above 2-1 searched faster through nonwords than through pseudowords and words, demonstrating a generalized effect of orthographic structure. Rumelhart's (1977) interactive model of reading proposed that readers utilize several knowledge sources (featural, graphic, phonemic, syntactic, orthographic, lexical, semantic) in word recognition. In contrast to earlier serial processing theories, the interactive model suggested that the knowledge sources were operating simultaneously, independent of level. As such, semantic processes constrain alternatives at the feature analytic level but are themselves constrained by feature analysis. In an extension of the interactive model designed to accommodate individual differences in reading fluency, Stanovich (1980) proposed that a deficit in any knowledge source would result in a heavier reliance on other knowledge sources, regardless of the level in the processing hierarchy. For example, the model predicts that readers who have difficulty with feature extraction may rely more heavily on other sources of information in word recognition. Stanovich found

Addressing issues in letter recognition.

Abstract: The present research strategy utilizes factorial designs, functional measurement, testing of mathematical models and strong inference in the study of letter perception. To test the viability of this framework, subjects judged a number of ambiguous letters, varying between Q and G, in both a rating and discrete choice task. The letters were created by varying features of openness in the oval and the obliqueness of the straight line. Experimental and theoretical tests on the results indicate that multiple sources of featural information simultaneously contribute to the perception of letters. The features provide continuous rather than discrete information to an integration process and the evaluation of the information provided by one feature is independent of the nature of the other features. The integration process results in the least ambiguous letter feature contributing the most to the perceptual judgment. A fuzzy logical model developed in other domains, such as speech perception, provides a good description of exactly these phenomena.

A Dynamic Programming Approach to Sequential Pattern Recognition

Abstract: This paper presents the dynamic programming approach to the design of optimal pattern recognition systems when the costs of feature measurements describing the pattern samples are of considerable importance. A multistage or sequential pattern classifier which requires, on the average, a substantially smaller number of feature measurements than that required by an equally reliable nonsequential classifier is defined and constructed through the method of recursive optimization. Two methods of reducing the dimensionality in computation are presented for the cases where the observed feature measurements are 1) statistically independent, and 2) Markov dependent. Both models, in general, provide a ready solution to the optimal sequential classification problem. A generalization in the design of optimal classifiers capable of selecting a best sequence of feature measurements is also discussed. Computer simulated experiments in character recognition are shown to illustrate the feasibility of this approach.

Application of pattern recognition to metal ion chemical ionization mass spectra

Abstract: : Pattern recognition methods were used to evaluate the information content of mass spectrometry data obtained using transition metal ions as an ionization source. Data sets consisting of the chemical ionization mass spectra for Fe+ and Y+ with 72 organics (representing the six classes alkane, alkene, ketone, aldehyde, ether, and alcohol) and 24 alkanes (representing the three subclasses linear, branched, and cyclic) were subjected to pattern recognition analysis using a k-nearest neighbor approach with feature weightings. The reactivities of Fe+ and Y+ toward the classes of compounds studied were characterized using classification accuracies as a measure of selectivity, and important chemical information was extracted from the raw data by empirical feature selection methods. A total recognition accuracy of 81% was obtained for the recognition of the six organic classes and 96% accuracy was obtained for the recognition of the three subclasses of alkanes. Keywords: Artificial intelligence; and Chemical ionization mass spectrometry.

The mental representation of faces: Spatial and temporal factors

Abstract: Three experiments are reported on the mental representation of faces with respect to the production of a face superiority effect. The effects of varying spatial position of the features and type of accompanying feature were investigated. Variations in attention to facial versus nonfacial features were considered by regression analyses, allowing an assessment of which was the more facelike of any two displays. Such regression analyses may have application to other recognition tasks if attention is divided between aspects of the display. Two further experiments explored the role of exposure duration in face superiority effects. The relationship of the results to models of face recognition is considered.

Voice recognition equipment

Abstract: PURPOSE: To improve the performance of recognition by judging the degree of contribution to recognition processing by the rate of partial simulatity to the sum of partial simularity after executing recognition processing once, resetting a weighted value by each feature based upon the judgement result and executing recognition processing for a succeeding input voice CONSTITUTION: An input pattern preparing part 10 inputs an input voice signal I, executes acoustic analysis processing, calculates a time series feature vector for the input voice based upon the analytical result, and outputs the calculated vector to a similarity calculation part 12 as an input pattern P The calculation part 12 calculates partial similarity by each feature and calculates the weighted sum of partial similarity obtained by referring the partial similarity and a feature-sorted weighted value W set up by each feature After executing recognition processing once, the rate of the partial similarity to the sum of partial similarity components judges, the degree of contribution to recognition processing, the weighted value W is reset and then recognition processing for a succeeding input voice is executed Since recognition processing optimum to a calculated feature value inherent in a specified speaker is executed, the performance of recognition is improved

A continuous speech recognition system based on knowledge engineering techniques

Abstract: The main objective of our research is to construct a continuous speech recognition system by using knowledge engineering techniques. The system simulates the behavior of a human expert who can recognize continuous speech by inspecting the trajectories of feature parameters such as formant frequencies. The expertise is embedded in the form of production rules. In the current implementation, we have 114 rules which are being updated. The introduction of production system to knowledge representation enables to cope with large amount of heuristics of speech recognition. Recognition rate of 85% was obtained for continuous speech of 30 second long uttered by three male adults. An environment for rule-base construction is also being developed.

Unsupervised speaker adaptation methods for vowel templates

Abstract: Four unsupervised speaker adaptation methods for vowel templates are described and evaluated There are two approaches to automatically obtaining information on vowel classification and location One is based on feature parameters and the other on the results of input speech recognition Here, the former is referred to as a bottom-up approach and the latter as a top-down one Two adaptation techniques are also presented The first is template selection from pre-stored sets and the second is template modification Combining these approaches and techniques, four adaptation methods are derived These four methods are evaluated in terms of spectral distortion and word recognition rate They are then compared considering performance, required calculation, rate of correctly used vowels, and type of input speech The results show that these methods work well and that the top-down approach is better than the bottom-up one They also show that the modification technique is better than the selection technique

Causal Uncertainty in the Identification of Environmental Sounds

Abstract: : This report is of an investigation into: (1) whether the recognition of an isolated environmental sound depends upon the number of different events that could cause the sound: (2) a method of quantifying the number of causal events; and (3) the cognitive processes that mediate the effect of multiple causation. Research in the past has focused on the acoustics of the sound in an attempt to determine which features the listener uses in recognition. However, it is well known that recognition is influenced by expectations, particularly about the number of alternatives. Three experiments on the effect of alternative causes are reported. The results of the first experiment replicated earlier results that the Hick-Hyman law applies to environmental sound identification and demonstrated the reliability of a measure of causal uncertainty. The measure is not a signal property in the usual sense. However, by reflecting the number of alternatives an individual considers in making a recognition judgment, it is a feature of a sound that is related to important aspects of recognition performance. The second experiment provided evidence toward the validity or this measure. Keywords: Auditory recognition; Auditory transients; Isolated sounds; Auditory sequences.

Toward a massively parallel system for word recognition

Abstract: This paper describes a massively parallel system for word recognition. Based on the connectionist network model adopted from cognitive science and artificial intelligence, the system consists of a large number of simple neuron-like processing units, or nodes, which represent words, phonetic segments, or phonetic features. The computation consists of constant updating of activation levels of all nodes, resulting from the excitatory links and inhibitory links between the nodes. Input to the system consists of frame-by-frame scores of similarity to a set of pre-defined spectral filters, which represents the set of phonetic segments necessary for distinguishing between words in the vocabulary. These similarity scores are combined into phonetic feature indexes for each frame of speech as input to the feature nodes in the network. A linguistic knowledge base is built into the network, allowing both data-driven processing and top-down prediction to cooperate or compete in working toward the correct lexical hypothesis.

Systolic architectures for connected speech recognition

Abstract: Systolic arrays for two connected speech recognition methods are presented. The first method is based on the dynamic time warping algorithm which is applied directly on acoustic feature patterns. The second method is the probabilistic matching algorithm which requires that the input sentence be preprocessed by a phonetic analyzer. It is shown that both methods may be implemented on either a two-dimensional or a linear systolic array. Advantages of each of these implementations are discussed. The architecture of a 12 000 transistors programmable NMOS prototype IC, which can be used as the basic processor of the probabilistic matching systolic arrays, is presented.

Three Layers Of Vector Outer Product Neural Networks For Optical Pattern Recognition

Abstract: A single homogeneous layer of neural network is reviewed. For optical computing, a vector outer product model of neural network is fully explored and is characterized to be quasi-linear (QL). The relationships among the hetero-associative memory [AM], the ill-posed inverse association (solved by annealing algorithm Boltzmann machine (BM)), and the symmetric interconnect [T] of Hopfield's model E(N) are found by applying Wiener's criterion to the output feature f and setting [EQUATION].

Voice-activated word processor with automatic learning for dynamic optimization of syllable-templates

Abstract: In this speaker-dependent recognition system, recognition is based on syllable template matching and each syllable has several templates. In the initial training for each speaker, 590 templates for 111 syllables are made, each including various contextual variations. The authors studied a learning method in which the syllable templates are automatically optimized. It is judged whether or not an input syllable should be learned according to the recent recognition condition. If it should be learned, the input syllable pattern replaces the template that contributes the least to recognition in the templates segmented from the same context and in the same syllable category. Automatic learning was evaluated on recognition of speech data obtained by reading Japanese sentences at a rate of about 4 to 5 syllables per second. The results over eight speakers showed an average syllable recognition accuracy of 71.0% without and 82.5% with automatic learning. Further, by increasing the maximum number of templates to 1024, it rose to 84.8%.

Fundamentals of image recognition (I)

Abstract: Feature extraction and matching are crucial to pattern recognition. However, there is no systematic presentation on these topics. The authors attempted to describe structures so that readers can review them broadly. Though, the content level is high, it is comprehensibly written, using many illustrated figures. This book treats two basic components of pattern recognition from both theoretical and and practical points of view.