Showing papers on "Feature (machine learning) published in 1969"

Methodologies of Pattern Recognition

Abstract: : The book presents the papers that were given at a conference on methodologies of pattern recognition. The titles include: implications of interactive graphic computers for pattern recognition methodology; statistical analysis as a tool to make patterns emerge from data; descriptive pattern analysis techniques, potentialities and problems; sequential pattern recognition systems; biological and mechanical pattern recognition; the automatic classification of fingerprints; cluster formation at various perceptual levels; recognition, machine recognition and statistical approaches; pattern recognition applied to the counting of nerve fiber cross-sections and water droplets; recognition by imitating the process of pattern generation; and the evaluation of the statistical classifier. (Author)

Learning in pattern recognition

Abstract: This paper is specifically concerned with the problem of inferring from a finite set of patterns the classification of an unknown pattern. A discussion of the general problems inherent in the concept of “learning” and “data reduction” are discussed from a standpoint of measurement selection for the general pattern recognition problem. A brief history of the existent work in empirical Bayes and compound sequential Bayes procedures will be presented. It is felt that these procedures are basically non-Bayesian, despite their names, and are therefore especially suited to problems arising in pattern recognition. Finally, a discussion is made of some nonparametric approaches to the problem of the classification of an unknown pattern when the only information on the underlying distributions associated with the various categories is that which can be obtained from a finite number of samples.

Feature selection for pattern recognition systems

Abstract: Publisher Summary The major problem of pattern recognition is essentially the discrimination of the input data between statistical populations via the search for features among members of a population. This chapter presents several approaches for the extraction of features in pattern recognition systems. The design of pattern recognition systems generally involves several major problem areas. The first problem is concerned with the representation of input data which can be measured from the objects of a pattern class. This is the sensing problem. The second problem is concerned with the selection of characteristic features or attributes from the received input data. This is often referred to as the feature extraction or selection problem. The third problem deals with the determination of optimum decision procedures which are needed in the process of identification and classification. This is the optimum decision problem. In solving the feature selection problem and the optimum decision problem, a set of parameters to be estimated and optimized is generally involved. This gives rise to the parameter estimation problem. The selection of features has been recognized as an important process in a pattern recognition system. When the complete set of discriminatory features for each pattern class can be determined from the measurement, the recognition and classification of the patterns will present no problem and automatic classification may be reduced to a simple matching procedure.

Descriptive pattern-analysis techniques: potentialities and problems

Abstract: Publisher Summary This chapter discusses the potentialities and problems of descriptive pattern-analysis techniques. The limitations inherent in the ability of the classical model to treat complex patterns are based on its inability to cope with what is thought of intuitively as the structure of a pattern and to concentrate its attention as necessary on the subpatterns whose relationships form this structure. The model is a global one, capable only of computing a set of properties defined on the whole input pattern and then making a choice based on them. That the model lends itself well to mathematical treatment is inadequate recompense for its failure in this respect. The classical pattern-classification model is a special (two-stage) case of pattern analysis, so defined, where the output of each feature extractor, usually a single number, is a very simple form of description of the input pattern, and the decision-making stage consolidates these descriptions into one description of the input pattern, a single number of specifying one of n categories. Thus, the entire effect is a mapping of a class of patterns into one of n possible descriptions.

Randomly Generated Nonlinear Transformations for Pattern Recognition

Abstract: In many mathematical and engineering problems the solution is simpler after a transformation has been applied. A general method is proposed to find suitable transformations for discrete data in information processing problems. The main feature of the method is random perturbation of the data subject to constraints which ensure that, in the transformed space, the problem is in some sense simpler and that the local structure of the data is preserved. An application of this technique to pattern recognition is discussed where a transformation is found for the feature space such that classes, which are not linearly separable in the original space, become so in the transformed space. The transformation considerably simplifies the problem and allows well-developed linear discriminant techniques to be applied. This application was implemented and tested with a number of examples which are described.

Recognition and action

Abstract: Publisher Summary The problem of pattern recognition cannot ultimately be separated from that of the organization of action: activity in view of ends. The concept of pattern is essentially relative; ambiguity can be removed only by talking of pattern-for-an-agent. The significance of a pattern for an agent lies in the conditional demands it makes on his organizing system. The best general solution to the pattern recognition problem may be to use feed-forward from feature filters to guide a self-moulding internal matching response at the appropriate level of the organizing system, using a combination of digital and analogue techniques. This chapter discusses the parallel between this process and the hypothetico-deductive procedure in science. It also highlights the advantages of null methods in pattern recognition, especially in connection with the problem of inspection.

Recognition, machine “recognition”, and statistical approaches

Abstract: Publisher Summary Although perception and recognition, and the logic of classification and statistical methodology have each been studied for decades, automatic pattern recognition is a new field. It is the recent emergence of computer related technology and needs that have led to serious attempts to synthesize various aspects of pattern recognition into machine recognition systems. In automatic pattern recognition, there is confusion between the goals of mechanical simulation of human perception and recognition, and machine systems to automate certain data-handling processes by performing a few pattern classification tasks. While in some situations these goals are related, they are nevertheless distinct. It has been long recognized that pattern recognition is an important aspect of almost all activities to which the appellation intelligent can be added confidently. This chapter discusses the relationship between recognition and classification, and between theory and application of statistical classification. Compound decision theory is the latest step in the evolution of the most general model in which to imbed statistical classification problems arising in recognition system design.

A Cloud Service Resource Classification Strategy Based on Feature Similarity

Abstract: There are now a vast array of heterogeneous cloud service resources, which makes it difficult to identify suitable services for the various types of cloud users. A classification of cloud service resources would help users find suitable cloud services more easily. We propose such a classification strategy, which has two parts. First, we improve the original naive Bayesian classification algorithm, designing a Bayesian classification algorithm based on feature similarity. Second, to improve the efficiency of the classification algorithm, we design a parallel programming model using the Hadoop platform. Simulation results show that the proposed classification strategy is feasible and effective, improving not only the resource classification accuracy but also greatly enhancing the processing efficiency for large-scale cloud service resources

Pattern recognition and a model of the brain

Abstract: The principal functions which characterize brain-like behaviour are considered, namely pattern recognition, pattern synthesis, memory and learning, and it is tentatively concluded that the brain can usefully be regarded as a pattern recognition machine. The primary features of the pattern recognition process, namely the concept of similarity, the use of context and the need for iterative signal analysis, are discussed. Finally, a model of a simple speech recognition machine which incorporates those aspects of brain processes which are relevant to such a machine is proposed.

The learning of parameters for generating compound characterizers for pattern recognition

Abstract: This paper presents and describes a pattern recognition program with a relatively simple and general basic structure upon which has been su­ perimposed a rather wide variety of techniques for learning, or self-organization. The program at­ tempts to generalize n-tuple approaches to pattern recognition, in which an n-tuple is a set of ind i ­ vidual cells or small pieces of patterns, and each n-tuple is said to characterize an input pattern when these pieces match i t , as specif ied. The program allows n-tuples to match when only some of their parts match, and it allows these parts to match even though they are not precisely positioned (See Uhr, 1969b, for some simple example programs). It further learns, in a variety of ways: It searches for good weights on its characterizers' implications, byre-weight­ ing as a function of feedback. It generates and discovers new characterizers (and can therefore begin with no characterizers at a l l ) , and discards characterizers that prove to be poor (See Uhr and Vossler, 1961, and Prather and Uhr, 1964). It also uses a set of characterizers of characterizers, to search for good parameter values that newlygenerated characterizers should have. A detailed f low-chartl ike "precis" descrip­ tion of the program is given, along with an ac­ tual l i s t ing . It is thus possible to examine ex­ actly what the program does, and how it does i t , and therefore to see how a wide variety of learning mechanisms have been implemented in a single pattern recognition program. But be­ cause it was coded in a "highlevel" patternmatching and l ist-processing language the pro­ gram runs too slowly for extensive tests to be practicable. Therefore only a brief l ist ing of output is given, to show that the program, works and begins to learn. Descriptors: Learning, self-organization, induc­ t ion , discovery, pattern recognition, learning to learn, n-tuple recognition, characterizing char­ acterizers.

Optical character recognition system including a matrix of scanned photosensitive elements

Abstract: A method of optical character recognition comprising utilizing a photohead having a matrix of photosensitive elements which are scanned successively and sampling the signals obtained during each scan for the purpose of detecting the presence of signals appertaining to a characteristic feature of one or more characters. The same feature may occur in several different characters but the features are chosen such that a different distinctive combination of features obtains for each character.

Potential functions in pattern classification algorithms

Abstract: Pattern classification algorithms using potential functions to construct discriminant functions from sample points set

Nonorthogonal projections for feature extraction in pattern recognition

Abstract: It is known that R linearly separable classes of multi-dimensional pattern vectors can always be represented in a feature space of at most R dimensions. An approach is developed which can frequently be used to find a non-orthogonal transformation to project the patterns into a feature space of considerably lower dimensionality. Examples involving classification of handwritten and printed digits are used to illustrate the technique.

Improvements relating to pattern recognition devices.

Abstract: 1,168,992. Pattern recognition. ELECTRIC & MUSICAL INDUSTRIES Ltd. 14 Dec., 1966 [25 Sept., 1965], No. 40902/65. Heading G4R. A pattern recognition device including stored representations of a plurality of different samples of each pattern to be recognized, the representations of different patterns being in different sub-groups of representations, comprises means for examining the stored representations to produce an indication that a feature is common to most of the stored representations and is not absent from most of each sub-group of representations and means for discarding that feature from storage in response to said indication. In the Figure, each row of store 1 stores indications (one bit per feature) of the features present in a standard versions of a possible character, for read-out and comparison in recognition device 2 with features of an unknown character. The three sub-groups consisting of rows 1-3, 4-6, 7-8 respectively relate to 1st, 2nd and 3rd possible characters. The rows are read out in turn, counters 6 totalling the numbers of occurrences of the respective features. Stores 8 receive the amounts the feature counts change during each sub-group. Each counter 6 then provides a "1" or "0" signal to a respective gate 9 if the count is above one threshold value or below a second (lower) value respectively. The stores 8 each provide a "1" or "0" signal to a gate 9 if the stored value is above or below a particular threshold value respectively. If all such signals from the counter 6 and stores 8 relating to a given feature are the same, the corresponding gate 9 prevents rewriting of the bits for that feature into store 1, by controlling inhibit gates 7 via mask store 10. The gate 9 also prevents that feature from being used in the recognition. The contents of store 1 may then be transferred to a smaller store. The store 1 may actually store a plurality of bits to represent each feature, feature recognition means at the store output producing the single bit referred to above from them.

Adaptive strategies of selecting feature subsets in pattern recognition

Abstract: A sequential decision model is discussed to design adaptive receptor for selecting feature subsets in recognition systems. Selection strategies are proposed to choose subsets of features based on the classification results fed back from the classifier. These strategies are adaptive in the sense that the selection is constantly guided by the feedback results in such a way as to maximize the long-run proportion of correct recognition. A character recognition experiment was carried out on a computer-simulated basis to demonstrate the feasibility of this model.

Classification of electro-encephalograms with pattern recognition algorithms.

Abstract: : The report proposes a mathematical model for making decisions about the condition of a subject from EEG date and algorithms for implementing the model. Pattern recognition methods are combined with the experience of a practicing electroencephalographer to balance the availability of mathematical models, computational feasibility, and experience. The aim of the model building is to produce a computationally feasible algorithm for a digital computer that generates a chart showing the condition of the subject as a function of time. The report gives preliminary results on feature extraction. In its present version, the pattern recognizer treats feature extraction and pattern classification distinctly and limits learning to transition probabilities of the Markov chain. The decision procedure that is outlined is applicable to any model that defines discrete states and permits Markovian movement between states. (Author)

A recognition system using probabilistic decisions based on extracted features

Abstract: The paper describes a pattern recognition system that has been simulated using a computer with an on-line camera input. The system is adaptive, using a training set of pictures together with the names or classes to which each picture belongs. The system uses an edge following technique for extracting features from the multilevel inputs. During the trainig mode, some of the descriptors derived from the extracted features are stored. Also, the system builds up statistics of the likelihood of a picture belonging to a given class given the presence of each incoming feature. During the test mode, a previously unseen set of pictures is used and features are extracted and compared with the stored descriptors. A sequential decision mechanism uses these comparisons and the likelihood statistics to produce responses corresponding to the assessed class of the input. Some preliminary experimental results are given.

Adaptive system of pattern recognition

Abstract: Publisher Summary The researches and developments on pattern recognition have recently resulted in the announcement of commercial reading machines one after another such as MICR and OCR, which can deal with both printed characters and hand written letters. This may seem to be a great advancement to the study of pattern recognition. However, the intrinsic ability of pattern recognition by human being has not been clarified yet. It is far from the situation that it is introduced to the mechanical recognition systems. The display systems for the output of the processed data from the information processing machines also have progressed very rapidly, for example, graphic displays, audio responses, and so on. The interface between man and the information processing machines has many troubles and problems, all of which are related to the ability of pattern recognition. The pattern recognition system in general is composed of the following elements: (1) input pattern, (2) environment, (3) parameter extraction, (4) decision algorithm, and (5) the adaptive or learning mechanism. This chapter describes the information processing system, the global decision mechanism for the adaptive situations, the selection strategy suitable for a lot of specific situations, and the pattern matching in actual implementations in several research works.

On sequential pattern recognition systems

Abstract: This paper is concerned with the application of sequential decision procedures to pattern recognition problems when the costs of feature measurements are of considerable importance. Three techniques are discussed, namely, the modified sequential probability ratio test (with time-varying stopping boundaries), the backward procedure using dynamic programming, and the nonparametric sequential ranking procedure. Two-class classification problems are first treated. Generalized procedures for multi-class problems are then given. In addition to classification problems, the backward procedure has also been applied to feature ordering and selection in sequential recognition processes. Computer simulated experiments in character recognition are presented to illustrate the proposed approaches.