Showing papers on "Feature selection published in 1983"

On the Selection of an Optimal Set of Indexes

Abstract: A problem of considerable interest in the design of a database is the selection of indexes. In this paper, we present a probabilistic model of transactions (queries, updates, insertions, and deletions) to a file. An evaluation function, which is based on the cost saving (in terms of the number of page accesses) attributable to the use of an index set, is then developed. The maximization of this function would yield an optimal set of indexes. Unfortunately, algorithms known to solve this maximization problem require an order of time exponential in the total number of attributes in the file. Consequently, we develop the theoretical basis which leads to an algorithm that obtains a near optimal solution to the index selection problem in polynomial time. The theoretical result consists of showing that the index selection problem can be solved by solving a properly chosen instance of the knapsack problem. A theoretical bound for the amount by which the solution obtained by this algorithm deviates from the true optimum is provided. This result is then interpreted in the light of evidence gathered through experiments.

Stepwise Location Model Choice in Mixed‐Variable Discrimination

Abstract: One practical drawback to the use of discrimination methods based on the location model for mixtures of discrete and continuous variables is that the smoothing techniques employed, and the subsequent estimation of error rates, limit fairly severely the allowable number of discrete variables. A backward elimination method of discrete variable selection is outlined in this paper. This can be used to identify a suitable, reduced location model for discriminant applications when the number of discrete variables is too large for direct use. It can also be used more traditionally as a variable selection procedure in discriminant analysis. Some examples are given.

Some criteria for variable selection in factor analysis

Abstract: Four criteria are proposed for variable selection in factor analysis. Three are introduced from the viewpoint to make the configurations of the true factor scores F and the estimated factor scores F(m) as close as possible. The remaining one comes from the maximization of the variance-covariance matrix due to regression of F on the variables X(m). The relationship among the four criteria and the generalized coefficient of determination (GCD) proposed by Yanai (1980) is discussed. The performances are investigated through the analyses of two sets of real data. As variable selection procedures we propose the forward selection procedure as well as the backward elimination procedure.

Feature Selection with Limited Training Samples

Abstract: A criterion which measures the quality of the estimates of the parameters of multivariate normal distributions for two class problems when limited number of samples are available is developed. This criterion predicts if the Hughes phenomenon occurs. The maximum number of features which does not degrade the accuracy of the classifier is then predicted. Experimental results regarding the Hughes phenomenon are included. Also presented is an example where the maximum number of features at each node in the binary tree classifier is predicted and compared with the maximum likelihood classifier. Index Terms-Training samples, multivariate normal distribution, Hughes phenomenon, feature selection, maximum likelihood classifier, bianry tree classifier.

Character recognizing method

Abstract: PURPOSE:To obtain a character recognizing device which has a high rate of character recognition by defining the covariance value matrix of feature values from a mean feature value and variance values obtained from extracted feature values as a dictionary and using it for decision making at a high speed. CONSTITUTION:A feature extraction part 1 extracts feature values from respective character data and stores them in a feature value storage part 2, and a feature value processing part 3 calculates the mean feature value and variance values from the feature values. A feature selection part 4 selects a feature effective to the representation of character data on the basis of the mean feature value and variance values. A covariance processing part 5 fetches the feature value corresponding to the feature selected by the feature selecting part 4 on the basis of some feature value in the feature value storage part 2 and the covariance matrix value is stored in a dictionary storage part 6 as a dictionary. The feature value from the feature extraction part 1, on the other hand, is supplied to a similarity calculation part 7, which calculates a similarity value from the signal from a dictionary storage part 6; and a candidate character storage part 8 determines and stores a candidate character category. A tournament processing part 9 generates a linear identification function from the covariance value matrix corresponding to the candidate character category name and obtains the decision result with the function.

Pattern recognition techniques in remote sensing data analysis

Abstract: This paper discusses several selected topics in decision-theoretic pattern recognition methods and introduces the syntactic approach to pattern recognition in remote sensing problems. The topics discussed are per-field classifications, mode estimation and sequential partitioning, feature selection and estimation of misclassification. The syntactic approach is introduced, and its application to remote sensing problems illustrated.

Unsupervised Feature Selection For Object Recognition

Abstract: A new method for selecting features for object recognition based on training data is proposed. This method avoids overspecifying or selecting too many features by using the criterion of minimal representation, which penalizes the representation complexity of features. The presented approach can be used to search for high level structural features such as relations or production rules.© (1983) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

A minicomputer based software system for the selection of optimal subsets of Thematic Mapper channels

Abstract: A software system has been developed and implemented on a minicomputer for feature selection based on two inter-dependent methods. The first is an enhancement of the traditional approach based on optimizing interclass average separabilities. The second is based on a Monte Carlo simulation of multispectral data and machine classification with subsequent estimation of classification accuracy as a function of channel subset. The two methods are mutually supportive - the first allows rapid screening whereas the second is based on the more solid theoretical foundation of maximizing classification accuracy.

Hybrid Optical-Digital Image Processing System For Pattern Recognition

Abstract: The coherent optical filtering techniques provide a general concept for the classification of patterns. This paper describes the design and testing of a hybrid optical-digital image processing system and the development of methods for a statistical expansion of the correlation signals. A conventional correlation signal intensity measurement is in most of the applications not sufficient. Six different algorithms for correlation signal evaluation are investigated. A feature reduction is achieved by multivariate analysis. For alpha-numeric patterns distored by binary random noise, rotation, scaling and shearing high classification results have been optained.

Parametric-histogramming technique for gross segmentation of color images of natural scenes (processing)

Abstract: This thesis describes a new systematic method for gross segmentation of color images of natural scenes. It has been developed within the context of the human visual system and mathematical pattern recognition theory. The eventual goal of the research is to integrate these two contexts to extract the visually distinct segments of an image which have vital importance for higher-level analysis or interpretation. A novel computational (pattern recognition) technique, called parametric-histogramming, is proposed in accordance with the human color perception and the Fisher criterion. This technique detects and isolates the image clusters efficiently and correctly using the 1-D parametric histograms of the L*,H('o),C* cylindrical coordinates of the (L*,a*,b*) - uniform color space in an unsupervised operation mode. In order to obtain the features most useful for a particular image, a new statistical-structural feature extraction method is devised in the form of a reference feature library containing several files. The underlying files tend to model the fundamental characteristics of uniformity, isolation, boundary, identation, texture, shadow and highlight patterns according to the grouping property of human eye and the Julesz conjecture. The dynamic operation characteristic of the feature selection phase is a significant property of the method. This type of operation, called dynamic feature space construction, enables the algorithm to select a particular feature space from the set of feature spaces so that image clusters in the selected space are more tractable and reliable.