Showing papers on "Feature vector published in 1987"

The acoustic-modeling problem in automatic speech recognition

Abstract: : This thesis examines the acoustic-modeling problem in automatic speech recognition from an information-theoretic point of view. This problem is to design a speech-recognition system which can extract from the speech waveform as much information as possible about the corresponding word sequence. The information extraction process is broken down into two steps: a signal processing step which converts a speech waveform into a sequence of information bearing acoustic feature vectors, and a step which models such a sequence. This thesis is primarily concerned with the use of hidden Markov models to model sequences of feature vectors which lie in a continuous space such as R sub N. It explores the trade-off between packing a lot of information into such sequences and being able to model them accurately. The difficulty of developing accurate models of continuous parameter sequences is addressed by investigating a method of parameter estimation which is specifically designed to cope with inaccurate modeling assumptions.

Optical character recognition by the method of moments

Abstract: An investigation of the use of two-dimensional moments as features for recognition has resulted in the development of a systematic method of character recognition. The method has been applied to six machine-printed fonts. Documents used to test the method contained 24 lines of alphanumeric characters. Before scanning a document to be processed, a training document having the same font must be scanned and stored in memory. Characters on the training document are isolated by contour tracing, and then the 2D moments of each character are computed and stored in a library of feature vectors. The document to be recognized is then scanned, and the 2D moments of its characters are compared with those in the library for classification. In this paper we present the selection of a set of moments that provide good discrimination between characters, the comparison of three classification schemes, the selection of a weighting vector that improves the classification performance, and a series of experiments to determine how the recognition rate is affected by the number of library feature vector sets. Recognition rates between 98.5% and 99.7% have been achieved for all fonts tested.

Integration of acoustic information in a large vocabulary word recognizer

Abstract: This paper proposes a new way of using vector quantization for improving recognition performance for a 60,000 word vocabulary speaker-trained isolated word recognizer using a phonemic Markov model approach to speech recognition. We show that we can effectively increase the codebook size by dividing the feature vector into two vectors of lower dimensionality, and then quantizing and training each vector separately. For a small codebook size, integration of the results of the two parameter vectors provides significant improvement in recognition performance as compared to the quantizing and training of the entire feature set together. Even for a codebook size as small as 64, the results obtained when using the new quantization procedure are quite close to those obtained when using Gaussian distribution of the parameter vectors.

Feature Selection for Automatic Classification of Non-Gaussian Data

Abstract: A computer-based technique for automatic selection of features for the classification of non-Gaussian data is presented The selection technique exploits interactive cluster finding and a modified branch and bound optimization of piecewise linear classifiers The technique first finds an efficient set of pairs of oppositely classified clusters to represent the data Then a zero-one implicit enumeration implements a branch and bound search for a good subset of features A test of the feature selection technique on multidimensional synthetic and real data yielded close-to-optimum, and in many cases optimum, subsets of features The real data consisted of a) 1284 12-dimensional feature vectors representing normal and abnormal breast tissue, extracted from X-ray mammograms, and b) 1060 30-dimensional feature vectors representing tanks and clutter in infrared video images

Normalization of speech by adaptive labelling

Abstract: In a speech processor system in which prototype vectors of speech are generated by an acoustic processor under reference noise and known ambient conditions and in which feature vectors of speech are generated during varying noise and other ambient and recording conditions, normalized vectors are generated to reflect the form the feature vectors would have if generated under the reference conditions. The normalized vectors are generated by: (a) applying an operator function A i to a set of feature vectors x occurring at or before time interval i to yield a normalized vector y i =A i (x); (b) determining a distance error vector E i by which the normalized vector is projectively moved toward the closest prototype vector to the normalized vector y i ; (c) up-dating the operator function for next time interval to correspond to the most recently determined distance error vector; and (d) incrementing i to the next time interval and repeating steps (a) through (d) wherein the feature vector corresponding to the incremented i value has the most recent up-dated operator function applied thereto. With successive time intervals, successive normalized vectors are generated based on a successively up-dated operator function. For each normalized vector, the closest prototype thereto is associated therewith. The string of normalized vectors or the string of associated prototypes (or respective label identifiers thereof) or both provide output from the acoustic processor.

Apparatus for identifying postage stamps

Abstract: An apparatus for identifying postage stamps automatically includes one or more scanners for detecting the various colors which are present at predetermined regions on the stamp proper and associated circuitry which receives the color signals and derives from them a feature vector which represents the color distribution over the scanned area. A plurality of feature vectors which define the color distribution associated with known stamps is stored in a memory. A comparator compares the feature vector of the stamp being examined to the prestored feature vectors and determines the identity of the stamps.

On Extensions to Fisher's Linear Discriminant Function

Abstract: This correspondence describes extensions to Fisher's linear discriminant function which allow both differences in class means and covariances to be systematically included in a process for feature reduction. It is shown how the Fukunaga-Koontz transform can be combined with Fisher's method to allow a reduction of feature space from many dimensions to two. Performance is seen to be superior in general to the Foley-Sammon method. The technique is developed to show how a new radius vector (or pair of radius vectors) can be combined with Fisher's vector to produce a classifier with even more power of discrimination. Illustrations of the technique show that good discrimination can be obtained even if there is considerable overlap of classes in any one projection.

A stochastic segment model for phoneme-based continuous speech recognition

Abstract: Developing accurate and robust phonetic models for the different speech sounds is a major challenge for high performance continuous speech recognition. In this paper, we introduce a new approach, called the stochastic segment model, for modelling a variable-length phonetic segment X, an L-long sequence of feature vectors. The stochastic segment model consists of 1) time-warping the variable-length segment X into a fixed-length segment Y called a resampled segment, and 2) a joint density function of the parameters of the resampled segment Y, which in this work is assumed Gaussian. In this paper, we describe the stochastic segment model, the recognition algorithm, and the iterative training algorithm for estimating segment models from continuous speech. For speaker-dependent continuous speech recognition, the segment model reduces the word error rate by one third over a hidden Markov phonetic model.

Visual pattern recognition in humans: I. Evidence for adaptive filtering

Abstract: We have investigated how observers learn to classify compound Gabor signals as a function of their differentiating frequency components. Performance appears to be consistent with decision processes based upon the least squares minimum distance classifier (LSMDC) operating over a cartesian feature space consisting of the real (even) and imaginary (odd) components of the signals. The LSMDC model assumes observers form prototype signals, or adaptive filters, for each signal class in the learing phase, and classify as a function of their degree of match to each prototype. The underlying matching process can be modelled in terms of cross-correlation between prototype images and the input sample.

Vector discrimination apparatus

Abstract: In a vector discrimination apparatus for performing discrimination such as class classification and recognition of a vector consisting of vector components corresponding to features of predetermined information, an optical pattern displayed on a display at a position corresponding to the size of the vector component is optically multiplied by a multiplier to form multiple images each having a substantially identical shape in the vicinities of various types of reference masks, and pattern matching is established between the multiple images and the various types of reference patterns formed on the reference masks. Discrimination such as class classification and recognition of the vector consisting of the vector components corresponding to the features of predetermined information such as a character or the like can be performed at a high speed, although the apparatus configuration is simple and low cost.

Optical associative processor for general linear transformations

Abstract: A new technique for the realization of general linear transformations using associative memories is described. An optical architecture for its implementation is also presented. A low-level feature space processor using this architecture is proposed. The processor is capable of recognizing and locating objects of various shapes and uses certain linear transformations in the feature space for distortion invariance.

Supervised Pattern Recognition Techniques In Quantitative Diagnostic Ultrasound

Abstract: The methods of statistical pattern recognition are well suited to the problems of in vivo ultrasonic tissue characterization. This paper describes supervised pattern recognition methods for selecting features for tissue classification, calculating decision boundaries within the selected feature space, and evaluating the performance. We address the considerations of dimensionality and feature size which are important in classification problems where the underlying probability distributions are not completely known. Examples are given for the detection of diffuse liver disease in the clinical environment.

Two-dimensional partially visible object recognition using efficient multidimensional range queries

Abstract: An important task in computer vision is the recognition of partially visible two-dimensional objects in a gray scale image. Recent works addressing this problem have attempted to match spatially local features from the image to features generated by models of the objects. However, many algorithms are less efficient than is possible. This is due primarily to insufficient attention being paid to the issues of reducing the data in features and feature matching. In this paper we discuss an algorithm that addresses both of these problems. Our algorithm uses the local shape of contour segments near critical points, represented in slope angle-arclength space (θ-s space), as the fundamental feature vectors. These fundamental feature vectors are further processed by projecting them onto a subspace of θ-s space that is obtained by applying the Karhunen-Loeve expansion to all critical points in the model set to obtain the final feature vectors. This allows the data needed to store the features to be reduced, while retaining nearly all their recognitive information. The resultant set of feature vectors from the image are matched to the model set using multidimensional range queries to a database of model feature vectors. The database is implemented using an efficient data-structure called a k-d tree. The entire recognition procedure for one image has complexity O(IlogI + IlogN), where I is the number of features in the image, and N is the number of model features. Experimental results showing our algorithm's performance on a number of test images are presented.

Position, Scale, and Rotation Invariant Target Recognition Using Range Imagery.

Abstract: : This thesis explores a new approach to the recognition of tactical targets using a multifunction laser radar sensor. Targets of interest were tanks, jeeps, and trucks. Doppler images were segmented and overlaided onto a relative range image. The resultant shapes were then transformed into a position, scale, and rotation invariant (PSRI) feature space. The classification processes used the correlation peak of the template PSRI space and the target PSRI space as features. Two classification methods were implemented: a classical distance measurement approach and a new biologically-based neural network multilayer perception architecture. Both methods demonstrated classification rates near 100% with a true rotation invariance demonstrated up to 20 degrees. Neural networks were shown to have a distinct advantage in a robust environment and when a figure of merit criteria was applied. A space domain correlation was developed using local normalization and multistage processing to locate and classify targets in high clutter and with partially occluded targets.

Character recognizing device

Abstract: PURPOSE:To simply learn a new category by producing a mapping matrix and recognizing dictionary for extraction of features of a discriminating method and a feature vector of each category via the K-L evolution of a single time. CONSTITUTION:An average initial vector of the learning character categories is produced from an initial vector where the feature values detected out of the pictures of learning categories. An inter-category covariance matrix is obtained from said average initial vector and the average initial vector of the categories registered before learning. Then an inter-normalized category covariance matrix is obtained from the product obtained between a normalized mapping matrix produced previously from an intra category covariance matrix and the inter-category covariance matrix. Then the proper vectors obtained by applying the K-L evolution to the inter-normalized category covariance matrix are arranged for production of an inter-category emphasis matrix which emphasizes the difference of categories. Then a matrix obtained from the product of the normalized mapping matrix and the inter-category emphasis matrix is used as a mapping matrix for decision/analysis. While a vector obtained from the product of the average initial vector of each category and the mapping matrix is used as a feature vector of a dictionary. Thus the learning is possible in a simple process.

Pattern recognition learning system

Abstract: PURPOSE:To execute a recognition processing efficiently by controlling the number of axes of a recognition dictionary in accordance with the number of dimensions of a feature which is extracted from an input pattern offered for the recognition processing CONSTITUTION:An input voice which has been inputted through a voice input part 1 is given to a feature extracting part 1, and a feature vector of its input voice pattern is extracted The feature vector of the input voice pattern which as been derived by the feature extracting part 2 is given to a recognizing part 3, and a learning part 4, at the time of a recognition processing of the input voice, and at the time of learning, respectively In the learning part 4, a generation processing of a recognition dictionary (standard pattern) conforming to a learning pattern is executed, and it is stored in a standard pattern dictionary memory 6 A control part 5 controls a learning method of the recognition dictionary in the learning part 4 from the number of the input voice pattern which have been used for generating the recognition dictionary in the learning part 4, and also controls the number of axes of the recognition dictionary used for the recognition processing of the input voice pattern in the recognizing part 3, in accordance with the number of dimensions of the feature vector of the input voice

Cofess--cooperative fuzzy expert systems for intelligent recognition

Abstract: COFESS is a pattern recognition system composed of three cooperating fuzzy expert systems (denoted by COFES1, COFES2 and COFES3) which utilizes fuzzy set theory and fuzzy logic in its decision making mechanisms. COFESS employs a recursion in the process of pattern recognition. Decisions related to the nature of the recognition need to be made along the way such as what feature to recognize next, etc. In order to solve this problem, an inference engine is constructed that examines a knowledge base and determines the next step in the recognition process. Another problem arises when we have to decide how one feature is related to the rest of the features that construct an object. Consider the problem of recognizing an object containing five identical squares--how can we prevent the system from recognizing the same square five times. To solve this problem (as well as other related problems) we defined two types of relations between features. The first type of relation determines the relative location of a feature with regard to other features and thus enables the system to distinguish between features. Moreover, by finding the area in which a certain element is expected to be found we are able to reduce the search space and increase the speed of the recognition process. The second type of relation is developed to help the system determine whether the feature recognized is indeed the feature that we intended to recognize. These are physical relations between the features (such as, how is the length of one feature related to the length of another feature, etc.), and are designed to help to distinguish between a feature and an accidental noise that resembles this feature. Upon successful localization of the designated area for recognition, a recognizer is activated to perform the actual pattern matching. Thus, the recognition of a feature involves four steps: (1) Deciding which element to recognize. (2) Finding the local area in which this element can be found. (3) Performing the pattern matching. (4) Checking whether or not this element is really the element which was expected to be recognized. (Abstract shortened with permission of author.)

An Improved Defect Classification Algorithm Based on Fuzzy Set Theory

Abstract: The characterization of defects in materials constitutes a major area of research emphasis. Characterization schemes often involve mapping of the signal onto an appropriate feature domain. Defects are usually classified by segmenting the feature space and identifying the segment in which the feature vector is located. As an example Udpa and Lord [1] map differential eddy current impedance plane signals on to the feature space using the Fourier Descriptor approach. Doctor and Harrington [2] use the Fisher Linear Discriminant method to identify elements of the feature vector that demonstrate a statistical correlation with the nature of the defect. Mucciardi [3] uses the Adaptive Learning Network to build the feature vector. In all these cases defect classification is typically accomplished by categorizing the mapped feature vectors using Pattern Recognition methods employing either distance or likelihood functions [4].

Generation of object descriptions from range data using feature extraction by demands

Abstract: A new method, called feature extraction by demands (FED), for generating an object description concurrently at different feature levels will be described. An object is described in terms of features which include points, surface patches, edges, corners, and surfaces. These features form a feature space which is the base used to decompose the feature extraction process into different levels. FED provides a method to generate partial descriptions about objects from partially processed range data at different feature levels. The partial descriptions become a feed-back to guide the feature extraction process to extract more detailed information from interesting areas which can then be used to refine the object description. Regions which are not perceived to contain useful infomation will be ignored in further processing. As a more complete object description is generated, FED converges from bottom-up image processing to top-down hypotheses verification to generate complete hierarchical object descriptions.

A feature based method for tracking mesoscale rainfall areas in radar images

Abstract: This paper presents a new feature-based matching algorithm for tracking mesoscale precipitation phenomena in radar image sequences. Distinct rainfall areas are identified in each image and characterized by a feature vector of shape descriptors, which provide a mathematical representation of the spatial characteristics of each identified area. Rainfall areas observed in consecutive images are matched by comparing the relative values of the features. Two match scoring algorithms are developed to generate the initial estimates of correct matches, which are then updated by likelihood measures based on relative location. The method is applied to mesoscale rainfall areas observed in sequences of radar-derived images of rainfall activity over Southwestern Ontario during the summers of 1980 and 1981.

Character recognition system

Abstract: PURPOSE:To increase a recognition speed by interrupting a matching processing relating to the type of a character which cannot be the candidate of an input character at an early time. CONSTITUTION:The distance between respective feature vectors and the feature vector of the dictionary pattern of the type of the character (k) is operated up to a high order Nth dimension by a matching part 24 and from the operated result, the standard deviation of the distance to the high order Nth dimension is obtained in a threshold decision part 28. The standard value is registered in a threshold table part 30 as the decision threshold of the character value (k). In the matching part 24, the matching distance (d) to the high order Nth dimension between the feature vector Fkn of the dictionary pattern of the type of the character (k) and the feature vector Yn of the input character is operated and the distance (d) is compared with the decision threshold Thk of the corresponding type of character (k) registered in the threshold table part 30 and decided. If d>Thk, since the type of the character of the input character may not be the current type of the character (k), the matching processing to the dictionary pattern is interrupted.

Pattern recognition device

Abstract: PURPOSE:To obtain a high recognition rate, by effectively utilizing a data at the time of producing a code book. CONSTITUTION:At the time of producing the code book, the system pattern of a large number of vectors is divided into the vector one by one, and clustering is performed at every vector, then, they are classified to several clusters, and the vector which represents the element of each cluster is stored as a code vector. Simultaneously, a weight matrix which applies weight on a distance by affecting the distribution state of each vector is found, and it is stored by corresponding to each code vector. And at the time of registering, the code vector resembled the vector exceedingly is found in the vector of the feature vector system of a standard pattern, and a quantization error is found. At the time of recognizing, the weight is applied on the distance by using the weight matrix corresponding to the code vector found at the time of producing the code book when distance calculation between the vector of an input pattern and the vector of the standard pattern is performed, and recognition is performed at a decision part 13 by using the distance.

Target Recognition Considerations

Abstract: This chapter discusses fundamental considerations in radar target recognition. The overall goal of the chapter is threefold: (1) to introduce some of the basic issues associated with the problem of target recognition, (2) to discuss and compare various target recognition techniques, and (3) to provide examples of applications of target recognition techniques, pointing out the limitations (and unrealized potential) of this technology. An extensive (but not exhaustive) list of references provides sources for further information on different aspects of the problem.

Threshold selection algorithm for extracting a uniform colour region in an image

Abstract: A new threshold selection algorithm for multidimensional feature space is proposed, and its applications to colour images are shown. This algorithm is based on tho recursive application of a simple 1-D threshold selection method (valley detection) to the conditional 1-D histograms. Although the performance of the algorithm such as convergence rate etc. is not yet evaluated, uniform colour regions are successfully extracted by the proposed algorithm for several colour images.

Clustering Analyser For Pattern Recognition

Abstract: Cluster analysis is a generic name for a variety of mathematical methods that can be used to classify a given data set. By using the cluster analysis the people try to understand a set of data and to reveal the structure of the data. Clustering technologies find very important applications in the disciplining of pattern recognition and image processing. They are very useful for unsupervised pattern classification and image segmentation. This paper presents a VLSI cluster analyser for implementing the squared-error clustering technique using extensive pipelining and parallel computation capabilities. The proposed cluster analyser could perform one pass of the squared-error algorithm (which includes finding the squared distances between every cluster center, assigning each pattern to its closest cluster center and recomputing the cluster centers) in 0(N+M+K) time units, where M is the dimension of the feature vector, N is the number of sample patterns, and K is the desired number of clusters. And it will need 0(N x M xK) time units, if a uniprocessor is used. The algorithm partition problem is also studied in this paper.