scispace - formally typeset
Search or ask a question

Showing papers on "Feature vector published in 1984"


Journal ArticleDOI
TL;DR: An electromyographic (EMG) signal pattern recognition system is constructed for real-time control of a prosthetic arm through precise identification of motion and speed command and the upper bound of probability of error and the average number of sample observations are investigated.
Abstract: An electromyographic (EMG) signal pattern recognition system is constructed for real-time control of a prosthetic arm through precise identification of motion and speed command. A probabilistic model of the EMG patterns is first formulated in the feature space of integral absolute value (IAV) to describe the relation between a command, represented by motion and speed variables, and location and shape of the corresponding pattern. The model provides the sample probability density function of pattern classes in the decision space of variance and zero crossings based on the relations between IAV, variance, and zero crossings established in this paper. Pattern classification is carried out through a multiclass sequential decision procedure designed with an emphasis on computational simplicity. The upper bound of probability of error and the average number of sample observations are investigated. Speed and motion predictions are used in conjunction with the decision procedure to enhance decision speed and reliability. A decomposition rule is formulated for the direct assignment of speed to each primitive motion involved in a combined motion. A learning procedure is also designed for the decision processor to adapt long-term pattern variation. Experimental results are discussed in the Appendix.

104 citations


PatentDOI
TL;DR: The work recognition system may be incorporated within an electronic device which is also equipped with speech synthesis capability such that the electronic device is able to recognize simple words as spoken thereto and to provide an audible comment via speech synthesis which is related to the spoken word.
Abstract: Speaker-independent word recognition method and system for identifying individual spoken words based upon an acoustically distinct vocabulary of a limited number of words. The word recognition system may employ memory storage associated with a microprocessor or microcomputer in which reference templates of digital speech data representative of a limited number of words comprising the word vocabulary are stored. The word recognition system accepts an input analog speech signal from a microphone as derived from a single word-voice command spoken by any speaker. The analog speech signal is directed to an energy measuring circuit and a zero-crossing detector for determining a sequence of feature vectors based upon the zero-crossing rate and energy measurements of the sampled analog speech signal. The sequence of feature vectors are then input to the microprocessor or microcomputer for individual comparison with the feature vectors included in each of the reference templates as stored in the memory portion of the microprocessor or microcomputer. Comparison of the sequence of feature vectors as determined from the input analog speech signal with the feature vectors included in the plurality of reference templates produces a cumulative cost profile for enabling logic circuitry within the microprocessor or microcomputer to make a decision as to the identity of the spoken word. The work recognition system may be incorporated within an electronic device which is also equipped with speech synthesis capability such that the electronic device is able to recognize simple words as spoken thereto and to provide an audible comment via speech synthesis which is related to the spoken word.

49 citations


Proceedings ArticleDOI
19 Mar 1984
TL;DR: A Hierarchical Vector Quantization scheme that can operate on "supervectors" of dimensionality in the hundreds of samples is introduced and Gain normalization and dynamic codebook allocation are used in coding both feature vectors and the final data subvectors.
Abstract: This paper introduces a Hierarchical Vector Quantization (HVQ) scheme that can operate on "supervectors" of dimensionality in the hundreds of samples. HVQ is based on a tree-structured decomposition of the original super-vector into a large number of low dimensional vectors. The supervector is partitioned into subvectors, the subvectors into minivectors and so on. The "glue" that links subvectors at one level to the parent vector at the next higher level is a feature vector that characterizes the correlation pattern of the parent vector and controls the quantization of lower level feature vectors and ultimately of the final descendant data vectors. Each component of a feature vector is a scalar parameter that partially describes a corresponding subvector. The paper presents a three level HVQ for which the feature vectors are based on subvector energies. Gain normalization and dynamic codebook allocation are used in coding both feature vectors and the final data subvectors. Simulation results demonstrate the effectiveness of HVQ for speech waveform coding at 9.6 and 16 Kb/s.

45 citations


Journal ArticleDOI
TL;DR: This paper proposes a general system approach applicable to the automatic inspection of textured material in which the input image is preprocessed in order to be independent of non-uniformities and a tone-to-texture transform is performed.

45 citations


Journal ArticleDOI
TL;DR: A class of registration algorithms that are reasonably efficient and robust for translational displacement has been considered and a new measure, namely, coefficient of variation, is defined to take into account effects of contrast and sharpness of the images.
Abstract: The automatic determination of local similarity between two images (image registration) is one of the most fundamental problems of image processing and pattern recognition. A class of registration algorithms that are reasonably efficient and robust for translational displacement has been considered to determine relative shift between reference and search images. Stochastic image models defined on a rectangular region of support are used to determine feature vectors associated with reference and search images. A new measure, namely, coefficient of variation, is defined to take into account effects of contrast and sharpness of the images. Based upon this measure, a computationally efficient two-stage algorithm is obtained by combining the image-model based algorithm with a template matching technique. Simulation results with several synthetic and real images are presented to evaluate the performance of the algorithms.

23 citations


Journal ArticleDOI
TL;DR: An investigation into the feasibility of applying pattern recognition concepts to the classification of metallic objects by their electromagnetic induction responses was performed and it is noted that the classifier extension developed provides a viable approach to classification of responses that very continuously with respect to a single parameter.
Abstract: An investigation into the feasibility of applying pattern recognition concepts to the classification of metallic objects by their electromagnetic induction responses was performed. The effect on the response of a limited set of steel spheroids due to various factors such as object shape, size, and orientation was examined and a pattern recognition scheme based on these results was proposed. Implementation of the scheme involved the development of a novel extension to the nearest mean vector type of classifier in which the concept of the class mean as a point in feature space was generalized to be a curve. The resultant pattern recognition scheme was tested on a representative test set which included 815 responses, corresponding to 104 variations in object and orientation. A success rate of greater than 96 percent was achieved. It is noted that the classifier extension developed provides a viable approach to classification of responses that very continuously with respect to a single parameter.

18 citations


Journal ArticleDOI
TL;DR: The performance using intensity and phase Fourier transform features and the performance in the presence of noise are studied and quantified for two different two-class pattern recognition data bases.
Abstract: Various feature extractors/classifiers for a hierarchical feature-space pattern recognition system are described. The system is intended to achieve multiclass distortion-invariant object identification. Although only a Fourier transform feature space is used, our basic hierarchical concepts, our theoretical analysis, and our general conclusions are applicable to other feature spaces. The performance using intensity and phase Fourier transform features and the performance in the presence of noise are studied and quantified for two different two-class pattern recognition data bases.

16 citations


PatentDOI
TL;DR: In this article, a speech signal recognition system compares the two-dimensional pattern (time sequence of feature vectors) of an unknown signal to prestored standard references patterns for recognition, thus forming a corresponding 2D comparison pattern of points of elemental Hamming distance differences.
Abstract: This speech signal recognition system compares the two-dimensionals pattern (time sequence of feature vectors) of an unknown signal to prestored standard references patterns for recognition, thus forming a corresponding two-dimensional comparison pattern of points of elemental Hamming distance differences. The sum of the pattern point distances is the similarity measure. To improve accuracy, partial patterns are selected (or "masked") and tested sequentially, and the point values weighted relative to their location within the mask. The mask may be rectangular or oblique.

15 citations


Patent
07 Feb 1984
TL;DR: In this article, a threshold for a rejection is determined by a threshold determining section 4 for every recognition word and stored in a threshold storage section 5 and a re-registration is performed if it is larger than the threshold of the beforehand set distance.
Abstract: PURPOSE: To improve the rejection performance when the voices, which are other than recognition objects, are inputted by automatically determining a recognition reject threshold value for every object voice at the time of the registration of recognition object voices and determining as to whether the recognition result is rejected or the recognition result is outputted by comparing the collation result with the corresponding threshold. CONSTITUTION: If a switch 8 selects a registration side, the speech signals inputted from a microphone are A/D converted by a feature extracting section 1 and stored in a standard pattern memory 3. These operations are conducted few times for every recognition word and a re-registration is performed if it is larger than the threshold of the beforehand set distance. Then, a threshold for a rejection is determined by a threshold determining section 4 for every recognition word and stored in a threshold storage section 5. If the switch 8 selects a recognition side, the feature vectors analyzed by the section 1 are stored in an input pattern memory 2, distances are calculated by a collating section 6, transmitted to a recognition discrimination section 7 and a rejection is conducted if the minimum distance exceeds the threshold for the rejection.

13 citations


Proceedings ArticleDOI
04 Dec 1984
TL;DR: A two-level feature extraction classifier using a geometrical-moment feature space is described for multi-class distortion-invariant pattern recognition.
Abstract: A two-level feature extraction classifier using a geometrical-moment feature space is described for multi-class distortion-invariant pattern recognition. The first-level classifier provides object class and aspect estimates using multi-class Fisher projections and optimized two-class Fisher projections in a hierarchical classifier. Aspect estimates are provided from ratios of the computed moments. The second-level classifier provides the final class estimate, distortion parameter estimates and the confidence of the estimates. Extensive test results on a ship image database are presented.

7 citations


Proceedings ArticleDOI
B. Kammerer1, W. Kupper, H. Lagger
01 Mar 1984
TL;DR: A single-word speech-recognition system based on autocorrelation feature vectors is presented, and in each case recognition rates are obtained better than 95% for a 250-word recognition system.
Abstract: In this paper a single-word speech-recognition system based on autocorrelation feature vectors is presented. Existing comparable systems use an extra data word for each coefficient of a feature vector. Normally for large vocabularies, vector quantization is performed in order to reduce the resulting large amount of data. Another way to reduce the storage needed is proposed by using a rough vector coefficient quantization instead of vector quantization. If, for example, 16 autocorrelation coefficients coded with two bits each are stored in one 32 bit data word, one obtains, besides an optimal use of the available storage, a good facility of computing a distance between pairs of feature vectors in a very fast way. A modified distance measure based on the cityblock distance is introduced. It only takes about 200 ns to compute one distance with the aid of appropriate programmed read only memories. If coefficient quantization is involved instead of vector quantization and the modified distance measure is used, there is no loss of accuracy. In each case we obtained recognition rates better than 95% for a 250-word recognition system.

Proceedings ArticleDOI
06 Feb 1984
TL;DR: A hierarchial multi-level feature-space pattern recognition system for distortion-invariant object identification with attention given to dimensionality reduction and the use of non-unitary transformations.
Abstract: A hierarchial multi-level feature-space pattern recognition system is described. Multi- class distortion-invariant object identification is the purpose of this study. Attention is given to dimensionality reduction (to simplify computations) and to the use of non-unitary transformations (to achieve discrimination). A Fourier transform feature space is used. However, our basic hierarchial concepts, our theoretical analysis, and our general conclu­ sions are applicable to other feature spaces. The use of intensity versus phase features is studied and the performance of our system in the presence of noise is studied. Quantitative experimental data on 2 two-class pattern recognition databases are provided. 1. INTRODUCTION Distortion-invariant multi-class pattern recognition is considered using a feature space. Feature extraction, dimensionality reduction, discrimination and classification are addressed. A simplified block diagram of our hierarchial pattern recognition system is shown in Figure 1. We begin with a Fourier transform feature space, since such a representation is well- known [1] to allow significant data compression. We extract the amplitude, phase or both from the Fourier transform plane. As the first dimensionality reduction technique, we wedge ring detector (WRD) sample the Fourier transform plane data [2]. This reduces the dimen­ sionality of the feature space to 64. Next, we compute the dominant eigenvectors of the WRD- sampled autocorrelation matrix. This reduced subspace is calculated using a Karhunen-Loeve (K-L) transformation [3] or implemented by new efficient techniques [4] for computing the dominant eigenvectors and eigenvalues of a large matrix. This completes the dimensionality reduction steps in our system. To provide discrimination, we employ two non-unitary trans­ formation: the Fukunaga-Koontz (F-K) [5] and the Foley-Sammon (F-S) [6]. Our classifier selects the best subspace (based on the probability of error) from the K-L, F-K and F-S feature vectors.

Journal ArticleDOI
TL;DR: This paper studies in theory, and gives a solution to the following concerns which may eventually be simultaneous: 1) obtain alternative classification decisions, ranked by some decreasing order of class membership probabilities; 2) imperfect teacher at the learning stage, or effects of labeling errors due to unsupervised learning by clustering; 3) noncooperative teacher, manipulating the a priori class probabilities.
Abstract: This paper studies in theory, and gives a solution to, the following concerns which may eventually be simultaneous: 1) obtain alternative classification decisions, ranked by some decreasing order of class membership probabilities; 2) imperfect teacher at the learning stage, or effects of labeling errors due to unsupervised learning by clustering; 3) noncooperative teacher, manipulating the a priori class probabilities; 4) unknown a priori class probabilities. These requirements are taken into account by considering a game between the recognition system and the teacher, in a game theoretical framework. Both players will ultimately select ``mixed strategies,'' which are probability distributions over the set of N alternative pattern classes, determined for each feature vector to be classified. This solution concept is interpreted in terms of the requirements 1)-4); numerical algorithms, as well as numerical examples are given with their solutions.

Patent
08 Feb 1984
TL;DR: In this article, the authors proposed a method to execute pattern recognition with high accuracy by decreasing a frame of a character pattern by small quantity when a character is rejected after pattern recognition based on a selected category and by retrying the pattern recoginition.
Abstract: PURPOSE: To execute pattern recognition with high accuracy by decreasing a frame of a character pattern by small quantity when a character is rejected after pattern recognition based on a selected category and by retrying the pattern recoginition. CONSTITUTION: When a feature vector X of an extracted unknown character pattern is located far from either relative positions C1 or C2 of a feature vector registered in a dictionary beforehand and decided to be rejected, a reasonable category is found in accordance with the following sequence: step 1: a parameter concerned with position coordinates of a vector X is changed by small quantity, thereby denoting a feature vector X', and it is approximated to a distance r k up to a CK. Step 2: When the distance between the feature vector X' and the CK will not approach by the step 1, it is rejected and the operation is stopped, and otherwise the operation is reset to the step 1, and the same process is conducted. When the distance r k is within a threshold and the operational frequency is below the set value, the CK will be the recognized result. COPYRIGHT: (C)1985,JPO&Japio

Book ChapterDOI
01 Jan 1984
TL;DR: In this paper, the Fourier descriptors are obtained by expanding the complex contour function in a Fourier series and incorporating functions of Fourier coefficients which are invariant under transformation of the trajectory into a feature vector.
Abstract: A technique involving the use of Fourier descriptors for characterizing impedance plane trajectories to facilitate defect classification is presented. The Fourier descriptors are obtained by expanding the complex contour function in a Fourier series. Functions of Fourier coefficients which are invariant under transformation of the trajectory are derived and incorporated into a feature vector. Defect classification is obtained by using the K-Means algorithm to cluster the feature vectors. The principal advantage of the approach lies in the ability to reconstruct the curve from the coefficients. Other advantages include the insensitivity of the descriptors to drift in the eddy current instrument as well as variations in probe speed. Experimental evidence attesting to the ability of the approach to discriminate between trajectories and hence identify defects is presented.

Patent
08 Oct 1984
TL;DR: In this article, a quantizing device converts a multilevel pattern to a binary pattern and applies it to a CPU, where the pattern is stored temporarily from the CPU into a memory and recognition processing such as character feature extraction, difference degree calculation, or the like are performed, and recognition results are outputted to an output device.
Abstract: PURPOSE:To improve the recognition rate in a short discrimination time even for mixed characters by using a weighting direction exponential histogram and a difference degree calculating method such as an artificial Maharanobis' distance/ artificial Bayes discriminating equation or the like to recognize characters. CONSTITUTION:A quantizing device 12 converts a multilevel pattern to a binary pattern and applies it to a CPU13. This pattern is stored temporarily from the CPU13 into a memory 14, and recognition processing such as character feature extraction, difference degree calculation, or the like are performed, and recognition results are outputted to an output device 15. A filter having a coefficient of weight is used for the character pattern to obtain a feature vector (weighting direction exponential histogram). Distances between preliminarily generated and stored standard patterns of respective character types and a character to be recognized are calculated. In an equation obtaining the Maharanobis' distance, inherent values of respective axes are used up to the K axis but inherent values are kept constant with respect to the (K+1) axis and following axes to calculate a degree of difference.

Proceedings ArticleDOI
J. Ackenhusen1
01 Mar 1984
TL;DR: The architecture of a single board processor for executing a variety of frame-by-frame connected pattern matching techniques which use dynamic time warping is described, and the connected dynamic time warp processor (CDTWP) will operate with existing hardware presently used for isolated word recognition.
Abstract: The architecture of a single board processor for executing a variety of frame-by-frame connected pattern matching techniques which use dynamic time warping is described. The connected dynamic time warp processor (CDTWP) will operate with existing hardware presently used for isolated word recognition. The CDTWP receives input in the form of a sequence of LPC-based feature vectors calculated from a spoken string of connected words. Each input vector, presented at a period of 15 msec, is compared with each frame of every reference template and the results are used to continuously update a hypothesized concatenation of reference templates that best matches the input. This comparison and update operation is completed for a given test frame before the next test frame arrives 15 msec later, and as a result, the CDTWP may be used to recognize the earlier portions of a connected string before the later portions have been spoken. The CDTWP is an experimental tool to examine a class of connected word recognition algorithms and architectures in real time. As a result, it is designed to be programmable and to fit within an existing word recognition system. The programmability and single board constraints limit the total number of reference frames to 455 (about 11 word templates). Current design work on a processor for isolated word dynamic time warping, the DTWP, suggests that a second generation CDTWP may handle 6000 reference frames (150 templates) by the use of existing nonprogrammable processing elements and addition of more template memory.

Journal ArticleDOI
TL;DR: This correspondence considers the problem of matching image data to a large library of objects when the image is distorted and demonstrates that, for classification purposes, distortions can be characterized by a small number of parameters.
Abstract: This correspondence considers the problem of matching image data to a large library of objects when the image is distorted. Two types of distortions are considered: blur-type, in which a transfer function is applied to Fourier components of the image, and scale-type, in which each Fourier component is mapped into another. The objects of the library are assumed to be normally distributed in an appropriate feature space. Approximate expressions are developed for classification error rates as a function of noise. The error rates they predict are compared with those from classification of artificial data, generated by a Gaussian random number generator, and with error rates from classification of actual data. It is demonstrated that, for classification purposes, distortions can be characterized by a small number of parameters.

Patent
17 Feb 1984
TL;DR: In this article, a non-linear mapping method was used to reduce in dimension multiple dimensional space into low dimensional space to visualize it, and executing correction such as deletion, addition, etc., of study data on the basis of human judgement.
Abstract: PURPOSE:To classify and process highly accurately objective data by using a non-linear mapping method, reducing in dimension multiple dimensional space into low dimensional space to visualize it, and executing correction such as deletion, addition, etc., of study data on the basis of a human judgement. CONSTITUTION:When a specific internal organ is automatically extracted from a nulear magnetism resonance NMR scanner picture, firstly known information such as proton density (d), vertical/horizontal transition times T1 and T2, etc., of the specific inner organs are imaged in the feature amount space. In this case, since the feature amount space is of three dimensional, the space is reduced in the two dimensional feature space by the non-linear matching method. In such a way, the low dimensional data thus reduced are set to 6, 7, and 8 by mankind corresponding the dicision area to the materials A, B, and C. Next, the applicability of the sample as for each distribution of A, B, and C is checked and the correction such as deletion and addition, etc., of study data. In such a way the decision area is decided at high accuracy, and highly accurate classification processing of the objective data is performed.

Journal ArticleDOI
01 Apr 1984
TL;DR: Considering the difficulties associated with this approach, decision tree design directly from a set of labelled samples is proposed in this paper, and the resulting decision trees are operationally very efficient and yield attractive classification accuracies.
Abstract: The design and operation of the minimum cost classifier, where the total cost is the sum of the measurement cost and the classification cost, is computationally complex. Noting the difficulties associated with this approach, decision tree design directly from a set of labelled samples is proposed in this paper. The feature space is first partitioned to transform the problem to one of discrete features. The resulting problem is solved by a dynamic programming algorithm over an explicitly ordered state space of all outcomes of all feature subsets. The solution procedure is very general and is applicable to any minimum cost pattern classification problem in which each feature has a finite number of outcomes. These techniques are applied to (i) voiced, unvoiced, and silence classification of speech, and (ii) spoken vowel recognition. The resulting decision trees are operationally very efficient and yield attractive classification accuracies.


04 Dec 1984
TL;DR: A face recognition system was developed, based on the principles of Cortical Thought Theory (CTT), recently proposed by Dr. Richard L. Routh, specifically for the difficult task of human face recognition.
Abstract: : A face recognition system was developed, based on the principles of Cortical Thought Theory (CTT), recently proposed by Dr. Richard L. Routh as his doctoral dissertation at the Air Force Institute of Technology. Routh tested the CTT architecture successfully for speech processing. In order to evaluate this architecture as a generic sensory information processing model, CTT was tested for visual processing, specifically for the difficult task of human face recognition. The CTT gestalt transformation maps a 2-dimensional images into a 2-D coordinate point. The present system extracts six sub-images from a contrast-expanded image, calculates the 2-D gestalt coordinates, and stores the information in a database. Statistics are then calculated on at least five prototypes processed for each person. Overall performance of different sub- windows on a face are also determined. An unidentified person is recognized by calculating the six gestalt feature vectors, and then finding the closest match to previously stored data. The computer generates an order list by closeness of match. Performance testing of the system yielded a reliability of 90%.

Patent
02 Oct 1984
TL;DR: In this paper, a plurality of speech feature vectors are generated from the time series of the speech feature parameter for the input speech pattern, by taking account of knowledge concerning the variation tendencies of speech patterns, and the learning (preparation) of a reference pattern vectors for speech recognition is carried out by the use of these feature vectors thus generated.
Abstract: In the learning method of reference pattern vectors for speech recognition in accordance with the present invention, a plurality of speech feature vectors are generated (block 20) from the time series of speech feature parameter for the input speech pattern, by taking account of knowledge concerning the variation tendencies of the speech patterns, and the learning (preparation) of a reference pattern vectors for speech recognition is carried out (block 22) by the use of these speech feature vectors thus generated. In particular, the method according to the present invention will become effective when it is combined with a statistical pattern recognition method that can absorb wide variations in the speech patterns.

01 Jan 1984
TL;DR: This dissertation presents a methodology that attempts to perform as much classification analysis as possible during the off-line learning process, while only a very small subset of the range data needs to be processed for the on-line recognition.
Abstract: During the past few years there have been great advances in the techniques of acquisition of range data from various sensing systems. The problem of speedily and reliably interpreting these range data for industrial object recognition is becoming critically important in the field of robotics and computer vision. This dissertation presents a methodology to meet such a challenge. It attempts to perform as much classification analysis as possible during the off-line learning process, while only a very small subset of the range data needs to be processed for the on-line recognition. It is assumed that there are only a small number of object types and the objects can be approximated by polyhedra. An adaptive matched filter is developed to extract an object's surface feature vectors using range measurement and surface normals. A sparse representation for each object category j is constructed for classification purpose in the form of a Rj-table of selected feature vectors. The approach, based on the concept of the generalized Hough Transform, classifies an object by examining the maximum votes which it receives from various Rj-tables. During the off-line learning phase an object tree is constructed for each prototype. The massive geometric data of a given prototype can be reordered through the tree traversal. An optimal selection rule is established for minimizing the misclassification probability. In this way the common feature vectors among object categories are removed as much as possible, while the distinctive feature vectors are selected in the Rj-table with statistical significance. Thus, it will result in a significant saving of the on-line processing time in comparison to the conventional three-dimensional scene segmentation approach. An experiment with simulated range images of nine categories demonstrated the success of the proposed methodology.

Patent
28 Jan 1984
TL;DR: In this article, a matching test using storage and input information is proposed to prevent busy state on the way of retrieval and to improve retrieving speed by dispersing storage information with high similarity into a storage space to store it in an information retrieving method.
Abstract: PURPOSE:To prevent busy state on the way of retrieval and to improve retrieving speed by dispersing storage information with high similarity into a storage space to store it in an information retrieving method by a matching test using storage and input information. CONSTITUTION:A characteristic vector obtained by decomposing a character to stroke and extracting from a reading character pattern is set up in a register 16. While counting up a counter 12 by a signal 18, a comparator 14 reads out dictionary feature vector successively from the leading address of a memory 10, discriminates rough matching at a [ I ] part, and when the matching is obtained, starts precise matching at a [II] part and continues the test also at the part [ I ]. If the matching is obtained by the [ I ] part before the completion of the test at the [II] part, busy state is generated. Therefore, the generation of the budy state is prevented by storing the dictionary feature vector with high similarity in separated addresses of the memory 10.

Book ChapterDOI
01 Jan 1984
TL;DR: The Hough transform technique was originally a method for detecting, in images, lines and other shapes characterisable by analytic functions, but has recently been extended to handle the correlation of 2D and 3D shapes which have no analytic description.
Abstract: The Hough transform technique was originally a method for detecting, in images, lines and other shapes characterisable by analytic functions. It has recently been extended (primarily by Ballard) to handle the correlation of 2D and 3D shapes which have no analytic description. One way of describing the technique is as a mapping from a spatially indexed feature space to a non-spatially indexed parameter space for the purpose of scene segmentation. The image segmentation technique of histogramming-then-thresholding can be used as an illustration. A very crude, and only under special circumstances successful, segmentation technique is to create a histogram of grey-level image intensity levels, ie, map the spatially indexed intensity values into a non-spatiaily indexed intensity grey level feature space. If the image originally consisted of an object and background of very different average reflectance properties, the histogram may have two clearly defined peaks. Simple thresholding at the minimum between them may suffice to segment figure from ground. Extending this idea to a colour feature space one can see that points contributing to peaks in the non-spatially indexed colour space may originate from significant spatially extended segments of the image, the blue of the sky, the green of the grass, the red roofs etc. Yet another example: consider the optic flow or instantaneous velocity image, grouping the points that are all moving in the same direction with the same velocity may be sufficient to segment out of the image the onrushing traffic from the background, etc. etc.

Proceedings ArticleDOI
01 Mar 1984
TL;DR: A new concept for examining shapes as vectors in a shape space is described and it is demonstrated that the feature vectors determined by this procedure are independent of size rotation, and displacement.
Abstract: A new concept for examining shapes as vectors in a shape space is described. The shape space is defined in terms of its properties and the importance of the independence of the size variable to the shape vectors defined on this shape space is stressed. Also, two theorems essential to the process of comparing partial shapes to the complete shape are given. A new method for detecting the points on a shape which appear to dominate visual perception is described. This method, called the Adaptive Line of Sight Method detects the dominant points of high curvature. The critical points, or the dominant points of the shape, which are determined by this method are based on a set of axes that are dependent on the shape itself. Therefore, the points determined are independent of size, rotation, or relative displacement. This concept is utilized to extract features from a shape. It is demonstrated that the feature vectors determined by this procedure are independent of size rotation, and displacement.

Patent
27 Nov 1984
TL;DR: In this paper, a character with high identification performance was identified by deciding whether relative relation between inclination information found for plural character stroke is satisfied with a prescribed condition or not, and when satisfied, detecting a graphic feature on a character pattern.
Abstract: PURPOSE:To identify a character with high identification performance by deciding whether relative relation between inclination information found for plural character stroke is satisfied with a prescribed condition or not, and when satisfied, detecting a graphic feature on a character pattern. CONSTITUTION:A feature extracting part 40 processes the contents of a pattern memory 31 while reading out the contents, and when scanning the character in plural directions from a mesh point of a pattern area, extracts feature vectors from graphic feature calculating values in respective directions is extracted to a feature vector storing part 47 and shape features indicating the appearance distribution of a shape code expressing the shape of the background white part for respective partial areas in the pattern area is extracted to a shape feature storing part 49. part 50, plural identification parts 51, 52 execute identification using feature vectors stored in the storing part 47 and identification using the shape features formed in the storing part 49 respectively and an overall deciding 53 determines the input character to output to an output terminal 70.