scispace - formally typeset
Search or ask a question

Showing papers on "Feature extraction published in 1983"


Journal ArticleDOI
TL;DR: This procedure, which requires no restrictions on the direction of motion, nor the location and shape of environmental objects, has been applied successfully to real-world image sequences from several different task domains.
Abstract: A procedure for processing real world image sequences produced by relative translational motion between a sensor and environmental objects is presented. In this procedure, the determination of the direction of sensor translation is effectively combined with the determination of the displacement of image features and environmental depth. It requires no restrictions on the direction of motion, nor the location and shape of environmental objects. It has been applied successfully to real-world image sequences from several different task domains. The processing consists of two basic steps: feature extraction and search. The feature extraction process picks out small image areas which may correspond to distinguishing parts of environmental objects. The direction of translational motion is then found by a search which determines the image displacement paths along which a measure of feature mismatch is minimized for a set of features. The correct direction of translation will minimize this error measure and also determine the corresponding image displacement paths for which the extracted features match well.

126 citations



Journal ArticleDOI
Potter1
TL;DR: The massively parallel processor's computing power and extreme flexibility will allow the development of new techniques for scene analysis - realtime scene analysis, for example, in which the sensor can interact with the scene as needed.
Abstract: A review of the massively parallel processor (MPP) is provided. The MPP, a single instruction, multiple data parallel computer with 16K processors being built for NASA by Goodyear Aerospace, can perform over six billion eight-bit adds and 1.8 billion eight-bit multiplies per second. Its SIMD architecture and immense computing power promise to make the MPP an extremely useful and exciting new tool for all types of pattern recognition and image processing applications. The SIMD parallelism can be used to directly calculate 16K statistical pattern recognition results simultaneously. Moreover, the 16K processors can be configured into a two-dimensional array to efficiently extract features from a 128x128 subimage in parallel. The parallel search capability of SIMD architectures can be used to search recognition and production rules in parallel, thus eliminating the need to sort them. This feature is particularly important if the rules are dynamically changing. Finally, the MPP's computing power and extreme flexibility will allow the development of new techniques for scene analysis - realtime scene analysis, for example, in which the sensor can interact with the scene as needed.

66 citations


Patent
28 Mar 1983
TL;DR: In this article, an apparatus and method for automatic character recognition is presented, which automatically locates and recognizes alphanumeric characters in a scene viewed by a raster-scan type of sensor.
Abstract: An apparatus and method for automatic character recognition is presented ch automatically locates and recognizes alphanumeric characters in a scene viewed by a raster-scan type of sensor. By this method the apparatus can search the entire scene and recognize up to seven alphanumeric characters in a bounded area of known size by the correlation technique in less than 1.5 seconds. The technique used first locates the bounded area and alphanumeric characters and then performs the recognition. Two different feature extraction methods are used to locate the characters in the scene. The recognition function is performed by correlation using an incoherent electrooptical processor.

55 citations


Journal ArticleDOI
TL;DR: A method of detecting blobs in images by building a succession of lower resolution images and looking for spots in these images, and it is possible to calculate thresholds in the low resolution image, and to apply those thresholds to the region of the original image corresponding to the spot.
Abstract: A method of detecting blobs in images is described. The method involves building a succession of lower resolution images and looking for spots in these images. A spot in a low resolution image corresponds to a distinguished compact region in a known position in the original image. Further, it is possible to calculate thresholds in the low resolution image, using very simple methods, and to apply those thresholds to the region of the original image corresponding to the spot. Examples are shown in which variations of the technique are applied to several images.

45 citations


Journal ArticleDOI
TL;DR: An optical processor that realizes a generalized chord transformation is described and the wedge-ring detector samples of an autocorrelation are shown to be the histograms of the chord distributions.
Abstract: An optical processor that realizes a generalized chord transformation is described. The wedge-ring detector samples of an autocorrelation are shown to be the histograms of the chord distributions. This dimensionality reduced set of features is used as the feature vector inputs for a Fisher linear classifier to determine the class of the input object independent of geometrical distortions. Initial discussions on the use of different classifiers, the polarity of the classifier’s output, and selection of the image training set are also advanced.

27 citations


Journal ArticleDOI
TL;DR: A class of matrix arithmetic networks is proposed for implementing the Foley-Sammon feature extraction algorithm and for generating linear discriminant vectors in pattern classification.
Abstract: In statistical methods for image processing and pattern classification, large-scale matrix computations are often performed over huge image data bases. A class of matrix arithmetic networks is proposed for implementing the Foley-Sammon feature extraction algorithm and for generating linear discriminant vectors in pattern classification. Such VLSI feature extractors and pattern classifiers are in high demand in real-time artificial intelligence applications. Performances of the proposed VLSI image analyzers are compared with conventional software approaches using a uniprocessor computer.

26 citations


Journal ArticleDOI
M. Kuhn1, H. Tomaschewski
TL;DR: Possibility for improving the recognition accuracy are investigated for a given feature extraction, which is based on a short term spectrum analysis by means of band-pass filtering and a method based on spectral change is investigated, alone and in combination with dynamic programming.
Abstract: For isolated word recognition, possibilities for improving the recognition accuracy are investigated for a given feature extraction, which is based on a short term spectrum analysis by means of band-pass filtering. A number of preprocessing steps are discussed, which are to be applied prior to time alignment via dynamic programming. These preprocessing steps include normalization of short term spectra with respect to the long term spectrum, amplitude normalization and spectral channel contour smoothing. For nonlinear time alignment, a method based on spectral change is investigated, alone and in combination with dynamic programming. The resulting distance measure is incorporated into pattern recognition schemes according to the minimum distance and nearest neighbor principles. The different processing steps are evaluated in a speaker dependent mode of operation separately for two vocabularies: the ten German digits and twelve major German airport city names. In comparison with the use of standard mean normalization of short term spectra and dynamic programming, the aforementioned techniques allow for a performance improvement in terms of error rate reduction by a factor of 3-5, while at the same time offering savings in computing time and reference memory requirements by a factor of 10 and 3, respectively.

26 citations


Proceedings ArticleDOI
01 Apr 1983
TL;DR: This paper proposes a new method for the recognition of consonant based on the Perceptron model, which is composed of the sensory, feature extraction, response and lateral inhibition layers.
Abstract: This paper proposes a new method for the recognition of consonant based on the Perceptron model. The recognition model is composed of the sensory, feature extraction, response and lateral inhibition layers. The recognition scores of 90.4% to 98.4% are obtained for unvoiced affricates, unvoiced plosives, unvoiced and voiced fricatives.

22 citations


Journal Article
TL;DR: Results of an experiment on Visual Evoked Response are presented, showing that through abidimensional analysis of the recorded data the resolution achievable in the localization of brain electrical activity can be increased to less than 1 cm.
Abstract: The aim of this work is to describe a system for the mono- and bi-dimensional analysis of brain electrical activity. The analysis was carried on either by visual inspection of mono- and bi-dimensional data, or by automatic feature extraction from the bidimensional data. Because of the importance of visual inspection for the analysis of experimental data, particular care was devoted to optimize the displayed data perceptually. For automatic screening of large amounts of data (and to allow long term studies of clinical records), statistical facilities were also provided. One purpose of the system was to develop image processing algorithms oriented toward biomedical images, that could be easily implemented on special purpose, low cost hardware, like VLSI or microcomputer arrays. This was possible because of the modularity of the larger part of bidimensional processing, such as interpolation and statistical analysis. Results of an experiment on Visual Evoked Response are presented, showing that through abidimensional analysis of the recorded data the resolution achievable in the localization of brain electrical activity can be increased to less than 1 cm.

16 citations


Proceedings ArticleDOI
27 Jun 1983
TL;DR: HEX is a general purpose geometric feature extractor with an integrated circuit layout emphasis that has the ability to do a simple or detailed extraction of circuit features depending on a user-provided sequence of instructions.
Abstract: HEX is a general purpose geometric feature extractor with an integrated circuit layout emphasis. It differs from previous extractors in that it is process-independent and has the ability to do a simple or detailed extraction of circuit features depending on a user-provided sequence of instructions. Internally the extraction process is carried out by an "instruction engine". The extractor produces an output graph showing feature connectivity and attaches extracted attributes to the feature nodes. In circuit applications the post-processed output is suitable for detailed simulation. HEX is currently in use at Siemens Corporation, West Germany.

Journal ArticleDOI
TL;DR: Two syntactic methods for the recognition of seismic waveforms are presented in this paper and both show equal recognition performance, but the nearest-neighbor rule is much faster in computation speed.
Abstract: Two syntactic methods for the recognition of seismic waveforms are presented in this paper. The seismic waveforms are represented by strings of primitives. Primitive extraction is based on cluster analysis. Finite-state grammars are inferred from the training samples. The nearest-neighbor decision rule and error-correcting finite-state parsers are used for pattern classification. While both show equal recognition performance, the nearest-neighbor rule is much faster in computation speed. The classification of real data for earthquake/explosion is presented as an application example.

Journal ArticleDOI
Minsoo Suk1, Ohyoung Song
TL;DR: A simple and efficient curvilinear feature extraction algorithm based on minimum spanning trees of edge points and closely related to Zahn's previous work on Gestalt clustering is described.
Abstract: A simple and efficient curvilinear feature extraction algorithm is described. The algorithm is based on minimum spanning trees of edge points and closely related to Zahn's previous work on Gestalt clustering. Examples drawn from real world images are shown to demonstrate the capabilities and applicabilities of the algorithm. Stimulating interest and inducing application of the algorithm in the areas of computer vision and image understanding are among the major objectives of the paper.

01 Feb 1983
TL;DR: It is shown that this problem can be viewed as the process of finding skeletons in a gray-scale image after observing that line detection does not necessarily depend on gradient information, but rather is approachable from the standpoint of measuring total intensity variation.
Abstract: : In this paper, the authors address a basic problem in machine perception: the tracing of "line-like" structures appearing in an image. It is shown that this problem can be viewed as the process of finding skeletons in a gray-scale image after observing the following: (1) that line detection does not necessarily depend on gradient information, but rather is approachable from the standpoint of measuring total intensity variation; and (2) that smoothing the original image produces an approximate distance transform. An effective technique for extracting the delineating skeletons from an image is presented, and examples of this approach using aerial, industrial, and radiographic imagery are shown.

Journal ArticleDOI
TL;DR: A shape analysis method for handwritten characters based on polygonal approximation and a recognition on the basis of parallel fuzzy labelling is introduced.

PatentDOI
TL;DR: In this paper, a speech recognition apparatus for raising the recognition speed, which is provided with first and second processors, is described, where the first processor mainly carries out feature extraction of the input speech, and the second one mainly carrying out comparison and distinction of feature extraction patterns with and from the reference patterns.
Abstract: This invention relates to a speech recognition apparatus for raising the recognition speed, which is provided with first and second processors, the first processor mainly carrying out feature extraction of the input speech, the second one mainly carrying out comparison and distinction of feature extraction patterns with and from the reference patterns, thereby processing the feature extraction hand comparison and distinction in an equilibrium manner and expecting the speech recognition at high speed. Another apparatus has been proposed which allows the first processor to carry out the comparison and distinction process in parts, thereby raising the recognition rate and performing the recognition at further high speed.

Proceedings ArticleDOI
14 Apr 1983
TL;DR: The results indicate that arbitrarily-shaped image regions can be well identified and clustered using as features their 2-D LPC parameters.
Abstract: This paper is concerned with the use of 2-D linear prediction for image segmentation. It begins with a brief summary of the mathematics involved in 2-D linear predictive analysis of arbitrarily-shaped regions. Then, it introduces a 2-D LPC distance measure based on the error residual of 2-D linear prediction. Finally, it describes how the above results can be applied to image segmentation using a simple cluster seeking algorithm. The results indicate that arbitrarily-shaped image regions can be well identified and clustered using as features their 2-D LPC parameters.

DOI
04 May 1983
TL;DR: The POPEYE system offers a range of functions including algorithms for preprocessing, feature extraction, image modeling, focusing, automatic pan, tilt, and zoom, interactive communication with other devices, and convenient user interaction.
Abstract: : A gray-level image processing system has been constructed to provide capability for inspection, object orientation, object classification,and interactive control tasks in an inexpensive, stand-alone system with moderate processing speed. The POPEYE system offers a range of functions including algorithms for preprocessing, feature extraction, image modeling, focusing, automatic pan, tilt, and zoom, interactive communication with other devices, and convenient user interaction. The host processor is a Motorola 68000 processor with Multibus communication between principal modules, an image data bus for acquisition and storage and a pipeline bus for image preprocessing and programmable transform operations. The software structure provides hierarchical control over multiple i/o devices, file management of system storage, an image management package and a vector package. Performance of the system is evaluated using convolution filters, adaptive modeling, histogram modification, and connectivity analysis. Cellular logic operations, piecewise gradient segmentation, automatic focusing, and adaptive spatial filtering examples are described in detail. The system is being applied to a number of practical industrial applications. (Author)

Patent
23 Jun 1983
TL;DR: In this article, the covariance value matrix of feature values from a mean feature value and variance values obtained from extracted feature values as a dictionary and using it for decision making at a high speed.
Abstract: PURPOSE:To obtain a character recognizing device which has a high rate of character recognition by defining the covariance value matrix of feature values from a mean feature value and variance values obtained from extracted feature values as a dictionary and using it for decision making at a high speed. CONSTITUTION:A feature extraction part 1 extracts feature values from respective character data and stores them in a feature value storage part 2, and a feature value processing part 3 calculates the mean feature value and variance values from the feature values. A feature selection part 4 selects a feature effective to the representation of character data on the basis of the mean feature value and variance values. A covariance processing part 5 fetches the feature value corresponding to the feature selected by the feature selecting part 4 on the basis of some feature value in the feature value storage part 2 and the covariance matrix value is stored in a dictionary storage part 6 as a dictionary. The feature value from the feature extraction part 1, on the other hand, is supplied to a similarity calculation part 7, which calculates a similarity value from the signal from a dictionary storage part 6; and a candidate character storage part 8 determines and stores a candidate character category. A tournament processing part 9 generates a linear identification function from the covariance value matrix corresponding to the candidate character category name and obtains the decision result with the function.

Journal ArticleDOI
TL;DR: An integrated or ‘weak’ solution approach to a distributed parameter system model for dynamic image sequences forms the basis for a feature extraction and motion estimation algorithm.

Book ChapterDOI
01 Jan 1983
TL;DR: A class-conditional log-likelihood functional, what leads to a least-squares procedure, is defined and the estimation of the shape change and motion parameters with a gradient algorithm is computed.
Abstract: In this paper, we propose an extraction method of some features, which describe dynamic properties of atmospheric disturbances. We identify a disturbance in a satellite image by means of a spirals model, knowing that the luminance model is constant levels plus noise. Thus, we define a class-conditional log-likelihood functional, what leads to a least-squares procedure. We compute the estimation of the shape change and motion parameters with a gradient algorithm. Such a processing could be an aid in interactive treatments and in the cloud archives development.

Patent
30 Sep 1983
TL;DR: In this paper, the authors proposed a method to improve the recognition rate, precision, and speed of recognition greatly as compared with conventional recognition using only a single binary pattern, and expanded the range of the recognition carried out on the binary patterns by recognizing plural binary patterns of the same object, and making a composite analysis.
Abstract: PURPOSE: To improve the recognition rate, precision, and speed of recognition greatly as compared with conventional recognition using only a single binary pattern, and expands the range of the recognition carried out on the binary patterns by recognizing plural binary patterns of the same object, and making a composite analysis CONSTITUTION: Binary coding circuits 2-A and 2-B have different threshold values set The ability to recognize patterns in simple shapes shown in figures deteriorates, specially, owing to a stains on the surface of an element, difference in reflection factor, and noise due to an image of a background part, so recognizing operation is performed by using two kinds of binary pattern having different brightness levels Those two kinds of binary pattern are recognized individually by using a recognizing method such as pattern matching and feature extraction, and the recognition results are ORed to identify the OR result as the object Thus, a high recognition rate is realized even for an object in the noise background and an object having a damage such as absence and a stain, and the recognition precision and time are improved COPYRIGHT: (C)1985,JPO&Japio

Journal ArticleDOI
TL;DR: In this article, complex Japanese characters were used to assess whether such a feature extraction interpretation could be generalized to identifying complex rotated symbols, and identification response times were also found to be constant across all non-standard orientations of the characters.

Patent
06 May 1983
TL;DR: In this article, a feature extraction part scans an input signal (picture signal) 10 to count frequencies of intersections of directions (vertical and horizontal), and writes the result in a feature pattern storage part 2.
Abstract: PURPOSE:To classify many kinds of characters through relatively simple processing by collating directional feature information on an input character with standard directional information, and then outputting characters with similarity. CONSTITUTION:A feature extraction part 1 scans an input signal (picture signal) 10 to count frequencies of intersections of directions (vertical and horizontal), and writes the result in a feature pattern storage part 2. A directional feature extraction part 3 reads the frequencies of intersection in a couple of prescribed directions out of the feature pattern storage part 2 successively. Further, the frequencies of appearance of the same contents in the coupled directions are counted as respective directional features with a processing to all contents of the feature pattern storage part 2, and pieces of feature information by the directions are inputted to a directional collation part 5. A storage part 4 for pieces of information by standard directions stores direction couples as feature information by the directions with regard to every kind of character, and its frequency, and sends them to the part 5.

Journal ArticleDOI
TL;DR: The fuzzy set theoretic approach has been combined with a heuristic approach for the recognition of handwritten uppercase english alphabets and numerals and a recognition accuracy better than 92 percent was achieved.

Patent
24 Mar 1983
TL;DR: In this paper, the reference value is made smaller for rescanning, and recess part presence (f) is checked; and when this existence ( f) is discriminated, character S is distinguished from number 5.
Abstract: PURPOSE:To improve the recognition rate or the recognition precision without increasing a required time much, by changing the reference in extraction of character features to scan characters again. CONSTITUTION:Characters on an original are converted photoelectrically by a scanner, and the obtained analog signal is subjected to A/D conversion and is written in a memory. The video signal written in the memory is subjected to positioning, namely, one-character segmentation. Features of the pattern described by segmented one-character components of the binary video signal are extracted. For example, alphabet S and number 5 are very like each other but S has a jump in the last, and character S is characterized by having a recess part Q in this jump part. Then, the reference value is made smaller for rescanning, and recess part presence (f) is checked; and when this existence (f) is discriminated, character S is distinguished from number 5. After changing the reference value in feature extraction, characters are discriminated and pertinent characters are extracted while referring to a dictionary, and thus, the increment of the time required for character recognition is restrained as much as possible.

Journal ArticleDOI
TL;DR: A central Motorola MC68000 processor controls several NEC μPD 7720 signal-processing chips as hardware subroutines to provide the necessary speed and flexibility in a real-time feature extractor.
Abstract: We describe the architecture of a real-time feature extractor used in the encoding of simple moving pictures into 10 kbit/s or less. A central Motorola MC68000 processor controls several NEC μPD 7720 signal-processing chips as hardware subroutines to provide the necessary speed and flexibility.

Patent
29 Mar 1983
TL;DR: In this article, a character deformed greatly is given rescanning indicating information and in this case the processing part 4 holds the answer in an answer correction processing part 6 as a temporary answer, and also instructs a rescanner control part 5 to vary a feature extraction threshold value which corresponds to whether the lower half loop is present or not.
Abstract: PURPOSE:To relieve erroneous recognition of a character deformed greatly by adding rescanning indication information corresponding to said character. CONSTITUTION:A character to be recognized has features extracted by a feature extraction part 2. A recognition processing part 4 collates the extracted features with individual feature information read out of a dictionary memory 3 successively. A character deformed greatly is given rescanning indicating information and in this case, the processing part 4 holds the answer in an answer correction processing part 6 as a temporary answer, and also instructs a rescanning control part 5 to vary a feature extraction threshold value which corresponds to whether, for example, the lower half loop is present or not. Consequently, the feature extraction part 2 performs feature extraction again on the basis of the varied threshold value. The answer correction processing part 6 obtains a real answer on the basis of plural supplied answers.


Proceedings ArticleDOI
26 Oct 1983
TL;DR: A software system for object recognition using map-guidance is reported on, three algorithms have been developed to find areally extended objects in a digital image.
Abstract: The analysis of images can take advantage of existing knowledge; this may be denoted as data-driven or knowledge-based image analysis. One example is the use of topographic maps in the study of aerial imagery. We report on a software system for object recognition using map-guidance. Three algorithms have been developed to find areally extended objects in a digital image. Results can be applied to compare a map data base with an image; to monitor changes; to geometric and radiometric rectification; to support classification with training areas and other tasks.© (1983) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.