Showing papers on "Feature vector published in 1991"

Face recognition using eigenfaces

Abstract: An approach to the detection and identification of human faces is presented, and a working, near-real-time face recognition system which tracks a subject's head and then recognizes the person by comparing characteristics of the face to those of known individuals is described. This approach treats face recognition as a two-dimensional recognition problem, taking advantage of the fact that faces are normally upright and thus may be described by a small set of 2-D characteristic views. Face images are projected onto a feature space ('face space') that best encodes the variation among known face images. The face space is defined by the 'eigenfaces', which are the eigenvectors of the set of faces; they do not necessarily correspond to isolated features such as eyes, ears, and noses. The framework provides the ability to learn to recognize new faces in an unsupervised manner. >

Nonlinear principal component analysis using autoassociative neural networks

Abstract: Nonlinear principal component analysis is a novel technique for multivariate data analysis, similar to the well-known method of principal component analysis. NLPCA, like PCA, is used to identify and remove correlations among problem variables as an aid to dimensionality reduction, visualization, and exploratory data analysis. While PCA identifies only linear correlations between variables, NLPCA uncovers both linear and nonlinear correlations, without restriction on the character of the nonlinearities present in the data. NLPCA operates by training a feedforward neural network to perform the identity mapping, where the network inputs are reproduced at the output layer. The network contains an internal “bottleneck” layer (containing fewer nodes than input or output layers), which forces the network to develop a compact representation of the input data, and two additional hidden layers. The NLPCA method is demonstrated using time-dependent, simulated batch reaction data. Results show that NLPCA successfully reduces dimensionality and produces a feature space map resembling the actual distribution of the underlying system parameters.

Acoustical and environmental robustness in automatic speech recognition

Abstract: This dissertation describes a number of algorithms developed to increase the robustness of automatic speech recognition systems with respect to changes in the environment. These algorithms attempt to improve the recognition accuracy of speech recognition systems when they are trained and tested in different acoustical environments, and when a desk-top microphone (rather than a close-talking microphone) is used for speech input. Without such processing, mismatches between training and testing conditions produce an unacceptable degradation in recognition accuracy. Two kinds of environmental variability are introduced by the use of desk-top microphones and different training and testing conditions: additive noise and spectral tilt introduced by linear filtering. An important attribute of the novel compensation algorithms described in this thesis is that they provide joint rather than independent compensation for these two types of degradation. Acoustical compensation is applied in our algorithms as an additive correction in the cepstral domain. This allows a higher degree of integration within SPHINX, the Carnegie Mellon speech recognition system, that uses the cepstrum as its feature vector. Therefore, these algorithms can be implemented very efficiently. Processing in many of these algorithms is based on instantaneous signal-to-noise ratio (SNR), as the appropriate compensation represents a form of noise suppression at low SNRs and spectral equalization at high SNRs. The compensation vectors for additive noise and spectral transformations are estimated by minimizing the differences between speech feature vectors obtained from a "standard" training corpus of speech and feature vectors that represent the current acoustical environment. In our work this is accomplished by minimizing the distortion of vector-quantized cepstra that are produced by the feature extraction module in SPHINX. In this dissertation we describe several algorithms including the SNR-Dependent Cepstral Normalization, (SDCN) and the Codeword-Dependent Cepstral Normalization (CDCN). With CDCN, the accuracy of SPHINX when trained on speech recorded with a close-talking microphone and tested on speech recorded with a desk-top microphone is essentially the same obtained when the system is trained and tested on speech from the desk-top microphone. An algorithm for frequency normalization has also been proposed in which the parameter of the bilinear transformation that is used by the signal-processing stage to produce frequency warping is adjusted for each new speaker and acoustical environment. The optimum value of this parameter is again chosen to minimize the vector-quantization distortion between the standard environment and the current one. In preliminary studies, use of this frequency normalization produced a moderate additional decrease in the observed error rate.

Gender recognition from speech. Part II: Fine analysis.

Abstract: The purpose of this research was to investigate the potential effectiveness of digital speech processing and pattern recognition techniques for the automatic recognition of gender from speech. In part I Coarse Analysis [K. Wu and D. G. Childers, J. Acoust. Soc. Am. 9 0 (1991)] various feature vectors and distance measures were examined to determine their appropriateness for recognizing a speaker’s gender from vowels, unvoiced fricatives, and voiced fricatives. One recognition scheme based on feature vectors extracted from vowels achieved 100% correct recognition of the speaker’s gender using a database of 52 speakers (27 male and 25 female). In this paper a detailed, fine analysis of the characteristics of vowels is performed, including formant frequencies, bandwidths, and amplitudes, as well as speaker fundamental frequency of voicing. The fine analysis used a pitch synchronous closed‐phase analysis technique. Detailed formant features, including frequencies, bandwidths, and amplitudes, were extracted by a closed‐phase weighted recursive least‐squares method that employed a variable forgetting factor, i.e., WRLS‐VFF. The electroglottograph signal was used to locate the closed‐phase portion of the speech signal. A two‐way statistical analysis of variance (ANOVA) was performed to test the differences between gender features. The relative importance of grouped vowel features was evaluated by a pattern recognition approach. Numerous interesting results were obtained, including the fact that the second formant frequency was a slightly better recognizer of gender than fundamental frequency, giving 98.1% versus 96.2% correct recognition, respectively. The statistical tests indicated that the spectra for female speakers had a steeper slope (or tilt) than that for males. The results suggest that redundant gender information was imbedded in the fundamental frequency and vocal tract resonance characteristics. The feature vectors for female voices were observed to have higher within‐group variations than those for male voices. The data in this study were also used to replicate portions of the Peterson and Barney [J. Acoust. Soc. Am. 2 4, 175–184 (1952)] study of vowels for male and female speakers.

Normalized image feature processing

Abstract: A method of processing an image including the steps of: locating within the image the position of at least one predetermined feature; extracting from the image data representing each feature; and calculating for each feature a feature vector representing the position of the image data of the feature in an N-dimensional space, such space being defined by a plurality of reference vectors each of which is an eigenvector of a training set of like features in which the image data of each feature is modified to normalize the shape of each feature thereby to reduce its deviation from a predetermined standard shape of the feature, which step is carried out before calculating the corresponding feature vector.

Feature-based visual servoing of an eye-in-hand robot with improved tracking performance

Abstract: A method is presented for using image features in servoing a robot manipulator. Specifically, the concept of a feature is mathematically defined, and the differential relationship between the robot motion and feature vector is derived in terms of a feature Jacobian matrix and its generalized inverse. The feature-based PID (proportional-integral-derivative) controller is established with three scalar gains and an n*n matrix. By means of various examples, the method of feature-based servoing of a robot proposed is proved to be very effective for conducting object-oriented robotic tasks. >

Invariance and neural nets

Abstract: Application of neural nets to invariant pattern recognition is considered. The authors study various techniques for obtaining this invariance with neural net classifiers and identify the invariant-feature technique as the most suitable for current neural classifiers. A novel formulation of invariance in terms of constraints on the feature values leads to a general method for transforming any given feature space so that it becomes invariant to specified transformations. A case study using range imagery is used to exemplify these ideas, and good performance is obtained. >

Preprocessing means for use in a pattern classification system

Abstract: A method for preprocessing reference feature vectors representing patterns in order to form, for each selected pattern class, collections of regions. A hierarchy of possibility regions is formed, wherein all reference feature vectors of a pattern class are contained in each level of the hierarchy of possibility regions associated with the pattern class. This hierarchy is later used to exclude a pattern class from consideration if a feature vector representing an unknown pattern is not contained in some level of its associated hierarchy of possibility regions. A collection of certainty regions is used, wherein no reference feature vector not of a pattern class is contained within any certainty region associated with the pattern class. The certainty regions are later used to classify a feature vector representing an unknown pattern as belonging to a pattern class. A collection of confidence regions is used to identify, although not with certainty, an unknown input pattern, and assign a confidence value indicating the relative confidence associated with the possibility that this unknown input pattern belongs to the pattern class.

Radial basis function networks for speaker recognition

Abstract: A speaker recognition system, using a modified form of feedforward neural network based on radial basis functions (RBFs), is presented. Each person to be recognized has his/her own neural model which is trained to recognise spectral feature vectors representative of his/her speech. Experimental results on a 40-speaker database indicate that the modified neural approach significantly outperforms both a standard multilayer perceptron and a vector quantization based system. The best performance for 4 digit test utterances is obtained from an RBF network with 384 RBF nodes in the hidden layer, given an 8% true talker rejection rate for a fixed 1% imposter acceptance rate. Additional advantages include a substantial reduction in training time over an MLP approach, and the ability to readily interpret the resulting model. >

Adaptive fast fuzzy clustering system

Abstract: A parallel processing computer system for clustering data points in continuous feature space by adaptively separating classes of patterns. The preferred embodiment for this massively parallel system includes preferably one computer processor per feature and requires a single a priori assumption of central tendency in the distributions defining the pattern classes. It advantageously exploits the presence of noise inherent in the data gathering to not only classify data points into clusters, but also measure the certainty of the classification for each data point, thereby identifying outliers and spurious data points. The system taught by the present invention is based upon the gaps between successive data values within single features. This single feature discrimination aspect is achieved by applying a minimax comparison involving gap lengths and locations of the largest and smallest gaps. Clustering may be performed in near-real-time on huge data spaces having unlimited numbers of features.

Classification of facial features for recognition

Abstract: A facial feature classification technique that independently captures both the geometric configuration and the image detail of a particular feature is described. The geometric configuration is first extracted by fitting a deformable template to the shape of the feature (for example, an eye) in the image. This information is then used to geometrically normalize the image in such a way that the feature in the image attains a standard shape. The normalized image of the facial feature is then classified in terms of a set of principal components previously obtained from a representative set of training images of similar features. This classification stage yields a representation vector which can be used for recognition matching of the feature in terms of image detail alone without the complication of changes in facial expression. Implementation of the system is described and results are given for its application to a set of test faces. These results show that features can be reliably recognized using the representation vectors obtained. >

Time delay neural network for printed and cursive handwritten character recognition

Abstract: A time delay neural network is defined having feature detection layers which are constrained for extracting features and subsampling a sequence of feature vectors input to the particular feature detection layer. Output from the network for both digit and uppercase letters is provided by an output classification layer which is fully connected to the final feature detection layer. Each feature vector relates to coordinate information about the original character preserved in a temporal order together with additional information related to the original character at the particular coordinate point. Such additional information may include local geometric information, local pen information, and phantom stroke coordinate information relating to connecting segments between the end point of one stroke and the beginning point of another stroke. The network is also defined to increase the number of feature elements in each feature vector from one feature detection layer to the next. That is, as the network is reducing its dependence on temporally related features, it is increasing its dependence on more features and more complex features.

A dynamical system approach to continuous speech recognition

Abstract: An dynamical system model is proposed for better representing the spectral dynamics of speech for recognition. It is assumed that the observed feature vectors of a phone segment are the output of a stochastic linear dynamical system, and two alternative assumptions regarding the relationship of the segment length and the evolution of the dynamics are considered. Training is equivalent to the identification of a stochastic linear system, and a nontraditional approach based on the estimate-maximize algorithm is followed. This model is evaluated on a phoneme classification task using the TIMIT database. It is shown that the classification performance obtained using the proposed model is significantly better than that obtained using either an independent-frame or a Gauss-Markov assumption on the observed frames. >

Speech recognition apparatus using neural network and fuzzy logic

Abstract: A speech recognition apparatus has: a speech input unit for inputting a speech; a speech analysis unit for analyzing the inputted speech to output the time series of a feature vector; a candidates selection unit for inputting the time series of a feature vector from the speech analysis unit to select a plurality of candidates of recognition result from the speech categories; and a discrimination processing unit for discriminating the selected candidates to obtain a final recognition result. The discrimination processing unit includes three components in the form of a pair generation unit for generating all of the two combinations of the n-number of candidates selected by said candidate selection unit a pair discrimination unit for discriminating which of the candidates of the combinations is more certain for each of all n C 2 -number of combinations (or pairs) on the basis of the extracted result of the acoustic feature intrinsic to each of said candidate speeches and a final decision unit for collecting all the pair discrimination results obtained from the pair discrimination unit for each of all the n C 2 -number of combinations (or pairs) to decide the final result. The pair discrimination unit handles the extracted result of the acoustic feature intrinsic to each of the candidate speeches as fuzzy information and accomplishes the discrimination processing on the basis of fuzzy logic algorithms, and the final decision unit accomplishes its collections on the basis of the fuzzy logic algorithms.

A Weighted Probabilistic Neural Network

Abstract: The Probabilistic Neural Network (PNN) algorithm represents the likelihood function of a given class as the sum of identical, isotropic Gaussians. In practice, PNN is often an excellent pattern classifier, outperforming other classifiers including backpropagation. However, it is not robust with respect to affine transformations of feature space, and this can lead to poor performance on certain data. We have derived an extension of PNN called Weighted PNN (WPNN) which compensates for this flaw by allowing anisotropic Gaussians, i.e. Gaussians whose covariance is not a multiple of the identity matrix. The covariance is optimized using a genetic algorithm, some interesting features of which are its redundant, logarithmic encoding and large population size. Experimental results validate our claims.

An hypothesized Wiener filtering approach to noisy speech recognition

Abstract: The problem of speech recognition in a noisy environment is addressed, in particular the mismatch problem originated when training a system in a clean environment and operating it in a noisy one. When measuring the similarity between a noisy test utterance and a list of clean templates, a correction process, based on a series of Wiener filters built using the hypothesized clean template, is applied to the feature vectors of the noisy word. The filtering process is optimized as a by-product of the dynamic programming algorithm of the scoring step. Tests were conducted on two databases, one in Hebrew and the second in Japanese, using additive white and car noise at different SNRs. The method shows a good performance and compares well with other methods proposed in the literature. >

Method for using a feed forward neural network to perform classification with highly biased data

Abstract: An artificial neural network detects points in feature space outside of a boundary determined by a set of sample data. The network is trained using pseudo data which compensates for the lack of original data representing "abnormal" or novel combinations of features. The training process is done iteratively using a net bias parameter to close the boundary around the sample data. When the neural net stabilizes, the training process is complete. Pseudo data is chosen using several disclosed methods.

Texture segmentation by clustering of Gabor feature vectors

Abstract: Approaches the texture segmentation problem by clustering feature vectors created from a Gabor transform data block. Given an N*N image, the authors compute 24 Gabor transforms using Gabor kernels with six orientations and four sizes. This results in a Gabor data block composed of N/sup 2/ feature vectors of length 24. The feature vectors are then grouped based on their distribution in the high-dimensional feature space. The authors hypothesize that the pixels in a given group have similar characteristics, and thus are part of the same texture. Experimental results for segmenting a synthetic railroad track image were encouraging; a clear-cut segmentation of the image was obtained. By clustering Gabor features, the authors were able to segment an image into regions of uniform texture without prior knowledge of the types of texture, or the frequency and orientation characteristics of these textures. The clustering algorithm is a modified Kohonen self-organizing feature map. >

A speaker-independent label coding apparatus

Abstract: The present invention is related to speech recognition and particularly to a new type of vector quantizer and a new vector quantization technique in which the error rate of associating a sound with an incoming speech signal is drastically reduced. To achieve this end, the present invention technique groups the feature vectors in a space into different prototypes at least two of which represent a class of sound. Each of the prototypes may in turn have a number of subclasses or partitions. Each of the prototypes and their subclasses may be assigned respective identifying values. To identify an incoming speech feature vector, at least one of the feature values of the incoming feature vector is compared with the different values of the respective prototypes, or the subclasses of the prototypes. The class of sound whose group of prototypes, or at least one of the prototypes, whose combined value most closely matches the value of the feature value of the feature vector is deemed to be the class corresponding to the feature vector. The feature vector is then labeled with the identifier associated with that class.

Velocity and acceleration features in speaker recognition

Abstract: The performance of dynamic features in automatic speaker recognition is examined. Second- and third-order regression analysis examining the performance of the associated feature sets independently, in combination, and in the presence of noise is included. It is shown that each regression order has a clear optimum. These are independent of the analysis order of the static feature from which the dynamic features are derived, and insensitive to low-level noise added to the test speech. It is also demonstrated that while the static feature gives the best individual performance, multiple linear combinations of feature sets based on regression analysis can reduce error rates. >

Textural and spectral features as an aid to cloud classification

Abstract: The problem of classifying clouds seen on meteorological satellite images into different types is one which requires the use of textural as well as spectral information. Since multi-spectral features are of prime importance, textural features must be considered as augmenting, rather than replacing, spectral measures. Several textural features are studied to determine their discriminating power across a number of cloud classes including those which have previously been found difficult to separate. Although several features in the frequency domain are tested they are found to be less useful than those in the spatial domain with only one exception. The specific features recommended for use in classification depend on the type of classification to be undertaken. Specifically, different features should be used for a multi-dimensional feature space analysis than for a binary-tree rule-based classification.

Using Projective Invariants for Constant Time Library Indexing in Model Based Vision

Abstract: Projectively invariant shape descriptors allow fast indexing into model libraries, because recognition proceeds without reference to object pose. This paper describes progress in building a large model based vision system which uses many projectively invariant descriptors. We give a brief account of these descriptors and then describe the recognition system, giving examples of the invariant techniques working on real images. We demonstrate the ease of model acquisition in our system, where models are generated directly from images. We demonstrate fast recognition without determining object pose or camera parameters.

Cloud detection based on texture segmentation by neural network methods

Abstract: A novel method to detect and recognize clouds from remote sensing images is introduced. The detection and recognition of clouds are based on textures. The images are partitioned into homogeneously textured regions, and the interpretation of those textures is based on a texture map. This map is created by means of artificial neural network methodology. The use of neural network methods makes it possible to apply an unsupervised learning paradigm to train the map continuously. The texture map is created by a self-organizing process of feature vectors. This is performed in an unsupervised way. The labeling is achieved by a supervised process. >

Learning polynomial functions by feature construction

Abstract: We present a method for learning higher-order polynomial functions from examples using linear regression and feature construction Regression is used on a set of training instances to produce a weight vector for a linear function over the feature set If this hypothesis is imperfect, a new feature is constructed by forming the product of the two features that most effectively predict the squared error of the current hypothesis The algorithm is then repeated In an extension to this method, the specific pair of features to combine is selected by measuring their joint ability to predict the hypothesis' error

Speech recognition apparatus of speaker adaptation type

Abstract: A speech recognition apparatus of the speaker adaptation type operates to recognize an inputted speech pattern produced by a particular speaker by using a reference pattern produced by a voice of a standard speaker. The speech recognition apparatus is adapted to the speech of the particular speaker by converting the reference pattern into a normalized pattern by a neural network unit, internal parameters of which are modified through a learning operation using a normalized feature vector of the training pattern produced by the voice of the particular speaker and normalized on the basis of the reference pattern, so that the neural netowrk unit provides an optimum output similar to the corresponding normalized feature vector of the training pattern. In the alternative, the speech recognition apparatus operates to recognize an inputted speech pattern by converting the inputted speech pattern into a normalized speech pattern by the neural network unit, internal parameters of which are modified through a learning operation using a feature vector of the reference pattern normalized on the basis of the training pattern, so that the neural network unit provides an optimum output similar to the corresponding normalized feature vector of the reference pattern and recognizing the normalized speech pattern according to the reference pattern.

A method of processing an image

Abstract: A method of processing an image comprising the steps of: locating within the image the position of at least one predetermined feature; extracting from said image data representing each said feature; and calculating for each feature a feature vector representing the position of the image data of the feature in an N-dimensional space, said space being defined by a plurality of reference vectors each of which is an eigenvector of a training set of like features in which the image data of each feature is modified to normalise the shape of each feature thereby to reduce its deviation from a predetermined standard shape of said feature, which step is carried out before calculating the corresponding feature vector.