Showing papers on "Feature (machine learning) published in 1980"

Attributed Grammar-A Tool for Combining Syntactic and Statistical Approaches to Pattern Recognition

Abstract: Attributed grammars are defned from the pattern recognidon point of view and shown to be useful for descriptions of syntactic stuctures as well as semantic attributes in primitives, subpatterns, and patterns. A pattern analysis system using attributed grammars Is proposed for pattern classification and description. This system extracts primitives and their attributes after preprocessing, performs syntax analysis of the resulting pattern representations, computes and extracts subpattern attributes for syntactically accepted patterns, and finally makes decisions according to the Bayes decision rule. Such a system uses a combination of syntactic and statistical pattern recognition techniques, as is demonstrated by illustrative examples and experimental results.

Pattern recognition by audio representation of multivariate analytical data

Abstract: We present an objective recipe, based on the statistical distribution of data on each axis in k-dimensional space, for audio display of multivariate analytical data. Each measurement in the data vector is translated into an independent property of sound. Scaling is performed in a continuous manner. The case for k less than or equal to 9 is treated here, but it should be possible to approach k = 20 with slight loss of resolution and with somewhat increased risk of non-orthogonality. A significant feature of such a presentation of analytical data is that good acoustical standards are available so that the trained ear can associate each sound with a good estimate of the value in the original measurement. Excellent results were obtained when this method was applied to the pattern recognition of a test data set. Advantages over visual (graphical) representation schemes are discussed.

Classification in well-defined and ill-defined categories: evidence for common processing strategies.

Abstract: Early work in perceptual and conceptual categorization assumed that categories had criterial features and that category membership could be determined by logical rules for the combination of features. More recent theories have assumed that categories have an ill-defined structure and have prosposed probabilistic or global similarity models for the verification of category membership. In the experiments reported here, several models of categorization were compared, using one set of categories having criterial features and another set having an ill-defined structure. Schematic faces were used as exemplars in both cases. Because many models depend on distance in a multidimensional space for their predictions, in Experiment 1 a multidimensional scaling study was performed using the faces of both sets as stimuli, In Experiment 2, subjects learned the category membership of faces for the categories having criterial features. After learning, reaction times for category verification and typicality judgments were obtained. Subjects also judged the similarity of pairs of faces. Since these categories had characteristic as well as defining features, it was possible to test the predictions of the feature comparison model (Smith et al.), which asserts that reaction times and typicalities are affected by characteristic features. Only weak support for this model was obtained. Instead, it appeared that subjects developed logical rules for the classification of faces. A characteristic feature affected reaction times only when it was part of the rule system devised by the subject. The procedure for Experiment 3 was like that for Experiment 2, but with ill-defined rather than well-defined categories. The obtained reaction times had high correlations with some of the models for ill-defined categories. However, subjects' performance could best be described as one of feature testing based on a logical rule system for classification. These experiments indicate that whether or not categories have criterial features, subjects attempt to develop a set of feature tests that allow for exemplar classification. Previous evidence supporting probabilistic or similarity models may be interpreted as resulting from subjects' use of the most efficient rules for classification and the averaging of responses for subjects using different sets of rules.

Recent Developments in Pattern Recognition

Abstract: Extensive research and development has taken place over the last twenty years in the areas of pattern recognition. Areas to which these disciplines have been applied include business (e.g., character recognition), medicine (diagnosis, abnormality detection), automation (robot vision and automatic inspection), military intelligence, communications (data compression, speech recognition), and many others. This paper presents a very brief survey of recent developments in basic pattern recognition techniques.

Pattern recognition of human brain electrical potentials

Abstract: Since brain electrical potentials (BEPs) are correlated with a variety of behavioral and clinical variables, especially tight experimental designs are necessary. Primary analysis, which usually consists of spectral analysis, linear prediction, or zero-cross detection, should match the time scale and dynamics of the states or processes being investigated. Nonneural contaminants must be removed from BEPs prior to computation of summary features. Principal components analysis, ad hoc methods, and stepwise discriminant analysis have been used to extract independent, intuitively appealing, and good-classifying features, respectively. Most pattern classification algorithms have been applied to BEPs including decision functions, trainable classification networks, distance functions, syntactic methods, and hybrids of the preceding. Because of its wide availability, most studies have used stepwise linear discriminant analysis.

An investigation of the use of dynamic time warping for word spotting and connected speech recognition

Abstract: Several variations on algorithms for dynamic time warping have been proposed for speech processing applications. In this paper two general algorithms that have been proposed for word spotting and connected word recognition are studied. These algorithms are called the fixed range method and the local minimum method. The characteristics and properties of these algorithms are discussed. It is shown that, in several simple performance evaluations, the local minimum method performed considerably better then the fixed range method. Explanations of this behavior are given and an optimized method of applying the local minimum algorithm to word spotting and connected word recognition is described.

Recognition of unaspirated plosives--A statistical approach

Abstract: In this paper the results of a study of the computer recognition of unaspirated plosives in commonly used polysyllabic words uttered by three different informants are presented. The onglide transitions of the first two formants and their durations have been found to be an effective set of features for the recognition of unaspirated plosives. The rates of transition of these two formants as a feature set have been found to be significantly inferior to the features mentioned earlier. The maximum likelihood method, under the assumption of a normal distribution for the feature set, provides an adequate tool for classification. The assumption of both intergroup and intragroup independence of the features reduces recognition scores. A prior knowledge of target vowels is found necessary for attaining reasonable efficiency. A prior knowledge of voicing manner improves classification efficiency to some extent. The physiological factors responsible for the variation of the recognition score for the various plosives are discussed. For labials and velars the recognition score is very high, nearly 90 percent. An attempt to correlate the dynamics of tongue-body motion with the variations in recognition scores has been made. Back vowels as targets have been found to give improved classification of the preceding consonants. A comparison of the result of machine recognition with those of published results on perception tests has been included. The results are found to be of the same order.

Topics in Statistical Pattern Recognition

Abstract: Pattern recognition, from the broadest viewpoint, is the study of how one puts abstract objects or patterns into categories in a simple reliable way. We have chosen three areas to review: Nonparametric Discrimination, Finite Memory Learning, and Pattern Complexity. We feel that these statistical areas will ultimately play a role in any global pattern recognition theory which evolves.

Progress report on pattern recognition

Abstract: A survey of the field of pattern recognition is presented in a manner broad enough not to be limited to the progress in recent years alone. In this review the emphasis is on the principal problems, methods and applications, and hence the omission of many details is unavoidable. Pattern recognition is discussed both intuitively and, more formally, as a many-to-one mapping. Several sources of variability, leading to many different possible representations of a certain pattern, are mentioned. All of these representations have to be classified as one pattern indicated by a certain class. An essential aspect of pattern recognition, relevant for this classification, is the selection of features by means of preprocessing the input data. Another essential aspect is the decision process based upon these features and leading to the classification. Several types of preprocessing are treated. Global and local transformations suitable for different types of filtering, decomposition and segmentation are indicated. A discussion of the three techniques used in pattern recognition, namely statistical, fuzzy and linguistic, is given.

Feature extraction and recognition of Japanese voiceless stop consonants by principal component analysis

Abstract: At the present time, how to extract acoustic features of voiceless stop consonants is oneof the most difficult problems remaining unsolved in the field of automatic speech recognition.This paper describes a study on characteristic features of the Japanese voicelessstop consonants /p/, /t/ and /k/. A multi-dimensional statistical analysis method isapplied to analyze their spectra. Analysis reveals that the principal characteristics discriminatingbetween them are reflected in the accumulated power from about 1kHz upto 5kHz and the existence of a spectral peak. It is also found that these feature parametersmake it possible to separate the voiceless stop consonants from each other. Experimentsof automatic recognition based on the maximum likelihood method are alsoperformed. They are carried out on condition that the following vowel is correctlyknown beforehand and that there are no errors in detection of the noise onset. It isfound that a recognition rate as high as about 97% can be attained for the training dataset of known utterances if the recognition algorithm designed to make the best use ofthe transition patterns of the feature parameters is adopted. It is also shown that theaccurate detection of the moment of noise burst is essential to attain a high recognition rate.

A speech recognition machine for connected words

Abstract: Construction and performance of a machine for recognizing spoken connected words are described. In this machine, a new method for connected word recognition, namely inverse dynamic programming (DP) matching, is adopted. Two kinds of DP matching techniques are used in the inverse DP matching, one of which is the usual DP matching and the other is matching performed in a time reverse mode, starting from the end of speech. Combining the similarities obtained by these two kinds of matching, the similarities between input speech and word sequences are computed. Also a technique for rejecting candidates is used in the machine to reduce computation amount. The machine performance is tested by 1400 samples of connected digits. The recognition rate of 99.3% is obtained.

Character recognition processing system

Abstract: PURPOSE:To recognize stably the pattern of complicated structure such as hand written Chinese character by extracting a pattern portion of the character with one stroke, thereafter combining the pattern portions to form the portion pattern and recognize the character. CONSTITUTION:The stroke portion on the stroke memory 2 is successively transmitted to the portion pattern memory 901 of the portion pattern identification portion 9 through the stroke transmission control portion 12 to constitute the part pattern of the character. Every time the feature of the part pattern is extracted by the feature extracting portion 902, the part pattern of the input character is identified in the part pattern judging portion 904 by referring to the part pattern identification lexicon 903. The identification of the part pattern is completed, so that this is informed to the stroke transmission control portion 12 and the result thereof is fed to the part pattern identification table 11 of general judging portion 10 successively. When all the part patterns identification are completed, the identification result 6 is outputted.

Syntactic Approach to Pattern Recognition

Abstract: Syntactic approach to pattern recognition is introduced. Special topics discussed include primitive selection and pattern grammars, syntactic recognition and error-correcting parsing, and clustering analysis for syntactic patterns.

Character recognition system

Abstract: PURPOSE:To recognize characters easily by agreement between a feature extraction bit pattrn and memory contents of the operator's dictionary, by loading previously a personal dictionary into an ORC unit to generate a dictionary and by reading a form corresponding to the operator and by sending the feature extraction bit pattern to the dictionary and by collating the feature extraction bit pattern with memory contents of the operator's dictionary. CONSTITUTION:Character samples A11, B12, and C13 are inserted to read mechanism 14 of OCR unit 20 individually, and features are extracted by feature extraction part 15, and feature groups are stored in a memory as bit patterns to generate dictionary 16. Dictionaries A17, B18 and C19 for respective persons are generated from bit patterns for respective persons and are stored in floppy disc device 24. Here, in case that respective form entering persons A, B and C process forms A21, B22 and C23 entered by themselves, dictionaries of respective persons are loaded previously into the memory in OCR unit 30 to generate dictionary 27, and forms 21...23 are read, and features are extracted in 26 to send bit patterns to dictionary 27, and the bit pattern is collated with the bit pattern group from unit 24, and thus, character can be recognized easily due to coincidence between them.

Character decision system of character recognition unit

Abstract: PURPOSE:To alternate between reduction in the number of unreadable characters and extreme reduction in error read by providing the recognizing decision part of a character recognition part with primary and secondary decision parts and by sending the output of either one by an external command CONSTITUTION:A character pattern from character pattern register 11 is scanned by feature extraction part 12 to extract various feature values (fj) (j=1-m), which are inputted to primary decision part 13 and secondary decision part 14 When a feature value satisfying category (ci) among feature values fj is detected, decision part 13 outputs primary recognition result (cfi) Decision part 14 extracts a feature value required to reduce extremely the error read of the category decided by decision part 13 and output (csi) With external terminal T' OFF, AND circuit 10i and OR circuit 30i output signal (cfi) and with terminal T' ON, output result (csi) of decision part 14 is outputted simultaneously; at this time, signal (csi) becomes output (ci) via AND circuit 20i and OR circuit 30i

An interactive pattern recognition laboratory (IPRL)

Abstract: This paper describes an interactive pattern recognition laboratory. The laboratory was designed for both research and teaching. For the researcher, it provides standard pattern recognition functions, a hierarchically organized pattern recognition data base, and a multidimensional graphic display capability. For the student it provides, in addition to the above capabilities, a vehicle for developing new pattern recognition algorithms. In addition to not having to develop support software, the student may compare the performance of his algorithms in the same environment as the existing ones.

Two Experiments on Statistical Image Segmentation.

Abstract: : Recently there has been a considerable research interest in applying statistical pattern recognition theory to image segmentation As the image is rich in statistical information, effective segmentation of images into meaningful parts can be performed by using statistical techniques In this report, we present segmentation results on infrared and reconnaissance images using two different statistical pattern recognition methods The first experiment is on the Alabama data base infrared images using the Fisher's linear discriminant analysis (1) To preserve the inter-pixel dependence as much as possible, measurements are taken in the form of a 3 x 3 matrix That is we are dealing with matrix measurements instead of vector measurements as typically considered in statistical pattern recognition (Author)

Onnline recognition processing system of handdwritten character

Abstract: PURPOSE:To make it possible to provide a sufficient recognition capability even for characters where curves are main constitution elements, by changing the number of feature points, where respective strokes of an input character are approximated, according to the number of strokes of the input character. CONSTITUTION:The character from character information input device 1 is normalized into a proper size by pre-processing unit 2, and the number of strokes of the character is discriminated by stroke discrimination unit 3. If the number of strokes is over 4, each stroke is approximated in three feature points by stroke three-point approximating unit 4. If the number of strokes is below 3 in Hiragana (Japanese syllabary), number, etc., each stroke is approximated in six five-divided feature points by stroke six-point approximating unit 5. Next, inter-pattern distances are calculated by inter-pattern distance calculation unit 6 in respect to the standard pattern stored in standard pattern storage unit 7. Next, the minimum value is detected from inter-pattern distances is detected by minimum distance detection unit 8, and category indicating the minimum value is recognized as an input pattern.

Multiistage classification processing system of character pattern

Abstract: PURPOSE:To classify and process character patterns efficiently with a high precision by extracting two complementary features included in character patterns and performing the multi-stage collation processing considering deformation of characters caused by difference of font. CONSTITUTION:Input character 1 is processed by outside contact frame detection circuit 4 and is inputted to storage circuit 5 after being read by scanning-type photoelectric converter 2. The output of storage circuit 5 has complemetary features of white and black extracted by white and black pattern generation circuits 9 and 8 in classification unit 6, and they are applied to collation circuits 10 and 14 successively and are compared with contents of feature pattern tables 11 and 12 of white and black of a known character, and classification result 7 is obtained. Collation circuit 10 compares a small number of feature vectors with one another to select candidate category 13 for finer classification in the next stage, and collation circuit 14 compares all vectors with one another in respect to this category.

Character*pattern recognizing system

Abstract: PURPOSE:To perform the recognition for the character and pattern through one reading action by memorizing the same picture into the picture memory in the form of the plural number of binary-coded picture information featuring different slice levels and then selecting the picture memory for recognition. CONSTITUTION:Same picture 3 is applied to level slicer 4 and then delivers the picture signals which is binary-coded by different slice levels. And the binary-coded signals are memorized in picture memories 5-1, 5-2 and 5-3 each. Furthermore, the contents of these memories are selected by switching unit 6 to be supplied to feature extracting circuit 7, and picture signals extracted through circuit 7 are recognized by decision circuit 8. In case the recognition is impossible, unit 6 is controlled by the control signals and via redecision control circuit 9. Thus the recognition is possible by the different slice levels memorized in other memories 5-1, 5-2 and 5-3 each. As a result, the recognition can be given assuredly for the character and pattern just through one reading action.

A pattern recognition model of voice-based personal verification systems for forensic applications

Abstract: -The success of automatic speaker recognition in laboratory environments suggests applications in forensic science for establishing the Identity of individuals on the basis of features extracted from speech. A theoretical model for such a verification scheme for continuous normaliy distributed featureIss developed. The three cases of using a) single feature, b)multipliendependent measurements of a single feature, and c)multpleindependent features are explored.The number iofndependent features needed for areliable personal identification is computed based on the theoretcal model and an expklatory study of some speech featues.

A model of a human recognition system with 'thinking'.

Abstract: The human system of pattern recognition is explored. This system utilizes the method of syntactic feature comparison, activation of the patterns containing the feature, and sorting among the activated patterns. Both the pattern recognition process without thinking and with thinking are discussed, and examples are given. Contrary to the usual concept, the more complicated pattern has a better chance to be recognized correctly.

Deductive approach to automatic recognition of Russian spoken sentences

Abstract: A deductive recognizing strategy, which requires a permanent linguistic activity at the highest levels (pragmatic, semantic, syntactic) is described. Semantically probable and grammatically correct sentences are continuously predicted and checked against the recognizing speech signal. They build successions of predicted phoneme standards processed in the multidimensional acoustic phonetic feature domain. The verification algorithm evaluates distances from each discrete 20 millisecond long interval of the speech signal to the corresponding predicted phoneme standards, thus converting the dynamic spectrogram of the message into the the phonemogram. Correct phoneme verification of 80% of words was obtained in a pilot study applying the deductive recognition strategy.

A model of pattern recognition by binary orthogonal transformation

Abstract: An information processing model of pattern perception has been constructed and applied to computer recognition of hand-written numerals. Patterns are picked up in the form of 32 × 32 binary array and orthogonally transformed by mutual additions and substructions among the elements. The Hadamard power spectra are defined as the feature variates to make distribution memories of the patterns which are called engram. Discriminations are performed through template matching by Euclid distance minimization method. A clusteing method has been introduced as the learning process to rewrite the engram. It is possible to identify the patterns correctly, not only for their position shift but also for their changes of parts or inclination to some extent. Multidimensional scaling method is used for the appreciation of effectiveness of the model.