scispace - formally typeset
Search or ask a question

Showing papers on "Feature extraction published in 1990"


Journal ArticleDOI
01 Apr 1990
TL;DR: The basic theory and applications of a set-theoretic approach to image analysis called mathematical morphology are reviewed in this article, where the concepts of mathematical morphology geometrical structure in signals are used to illuminate the ways that morphological systems can enrich the theory and application of multidimensional signal processing.
Abstract: The basic theory and applications of a set-theoretic approach to image analysis called mathematical morphology are reviewed. The goals are to show how the concepts of mathematical morphology geometrical structure in signals to illuminate the ways that morphological systems can enrich the theory and applications of multidimensional signal processing. The topics covered include: applications to nonlinear filtering (morphological and rank-order filters, multiscale smoothing, morphological sampling, and morphological correlation); applications to image analysis (feature extraction, shape representation and description, size distributions, and fractals); and representation theorems, which shows how a large class of nonlinear and linear signal operators can be realized as a combination of simple morphological operations. >

336 citations


Journal ArticleDOI
TL;DR: Recognition experiments with a prototype system for a variety of complex printed documents shows that the proposed system is capable of reading different types of printed documents at an accuracy rate of 94.8–97.2%.

258 citations


Journal ArticleDOI
TL;DR: Results obtained show that direct feature statistics such as the Bhattacharyya distance are not appropriate evaluation criteria if texture features are used for image segmentation, and that the Haralick, Laws and Unser methods gave best overall results.

228 citations


Proceedings ArticleDOI
17 Jun 1990
TL;DR: The proposal of G. Cottrell et al. (1987) that their image compression network might be used to extract image features for pattern recognition automatically, is tested by training a neural network to compress 64 face images, spanning 11 subjects, and 13 nonface images.
Abstract: The proposal of G. Cottrell et al. (1987) that their image compression network might be used to extract image features for pattern recognition automatically, is tested by training a neural network to compress 64 face images, spanning 11 subjects, and 13 nonface images. Features extracted in this manner (the output of the hidden units) are given as input to a one-layer network trained to distinguish faces from nonfaces and to attach a name and sex to the face images. The network successfully recognizes new images of familiar faces, categorizes novel images as to their `faceness' and, to a great extent, gender, and exhibits continued accuracy over a considerable range of partial or shifted input

162 citations


Patent
30 Apr 1990
TL;DR: In this article, the authors proposed a layered network having several layers of constrained feature detection, where each layer of the network includes a plurality of feature maps and a corresponding plurality of reduction maps, each feature reduction map is connected to only one constrained feature map in the same layer for undersampling that constrained feature maps.
Abstract: Highly accurate, reliable optical character recognition is afforded by a layered network having several layers of constrained feature detection wherein each layer of constrained feature detection includes a plurality of constrained feature maps and a corresponding plurality of feature reduction maps. Each feature reduction map is connected to only one constrained feature map in the same layer for undersampling that constrained feature map. Units in each constrained feature map of the first constrained feature detection layer respond as a function of a corresponding kernel and of different portions of the pixel image of the character captured in a receptive field associated with the unit. Units in each feature map of the second constrained feature detection layer respond as a function of a corresponding kernel and of different portions of an individual feature reduction map or a combination of several feature reduction maps in the first constrained feature detection layer as captured in a receptive field of the unit. The feature reduction maps of the second constrained feature detection layer are fully connected to each unit in the final character classification layer. Kernels are automatically learned by constrained back propagation during network initialization or training.

123 citations


Journal ArticleDOI
TL;DR: A state of the art review of feature recognition techniques developed and feature based design as an alternative are presented and process planning systems developed using both approaches are presented.
Abstract: The quest for completely automated process planning systems has exposed the lack of techniques capable of automatically understanding the stored CAD models in a manner suitable for process planning. Most current generations of process planning systems have used the ability of humans to translate the part drawing requirements into a form suitable for computer aided process planing. Recently, research advances have been made to improve the understanding of computer stored 3-D part models. The two approaches used are feature recognition and feature based design. This paper presents a state of the art review of feature recognition techniques developed and presents feature based design as an alternative. Process planning systems developed using both approaches are presented.

107 citations


Journal ArticleDOI
TL;DR: This work extends automatic multivariate statistical classification techniques to find classes of pixels or features in the images (or other n-dimensional signals) that exhibit a homogeneous statistical behavior throughout the data set.
Abstract: Over the past few years, automatic multivariate statistical classification techniques have successfully been used for analyzing large and noisy electron-microscopic data sets. After the raw data are compressed with eigenvector–eigenvalue procedures, classes of images are formed, using unsupervised clustering procedures. The classes elucidate even subtle differences existing within the data set. Here we extend these methods to find classes of pixels or features in the images (or other n-dimensional signals) that exhibit a homogeneous statistical behavior throughout the data set. This feature extraction—itself a form of data compression—is mathematically entirely symmetric to the determination of the image classes and also serves the purpose of revealing the information present in the set of input images. The properties of simultaneous representations of the image-space and feature-space data onto the same two-dimensional map are discussed in relation to the metrics used in both spaces. Model data are used to illustrate the basic ideas.

63 citations


Proceedings ArticleDOI
16 Jun 1990
TL;DR: An algorithm to normalize the skew of document images is proposed, which shows that when graphical elements are included in the documents in addition to printed characters, the accuracy deteriorates to 0.2 degrees.
Abstract: An algorithm to normalize the skew of document images is proposed. The skew angle is detected in two stages. In the first stage, connected regions in an image are extracted and some feature parameters are extracted for each region. In the second stage, the Hough transform is calculated for the parameters, and the angle which gives the minimum of the transform is estimated as the skew angle. In experiments using CCITT standard documents, a detection accuracy of less than 0.1 degrees is obtained for printed documents. When graphical elements are included in the documents in addition to printed characters, the accuracy deteriorates to 0.2 degrees . >

62 citations


Patent
15 Jun 1990
TL;DR: In this paper, a method and algorithm for rapidly quantifying phagocytic functions using computer image analysis (CIA) of video light microscopic images is presented, which involves sequential acquisition of bright field or phase contrast and epi-fluorescence video microscopic images of respective field, addition of the images, decision making, object referencing, morphological feature extraction, arithmetic operations, and statistical analysis.
Abstract: The present invention is embodied in a method and algorithm for rapidly quantifying phagocytic functions using computer image analysis (CIA) of video light microscopic images. The method and algorithm involve sequential acquisition of bright field or phase contrast and epi-fluorescence video microscopic images of respective field, addition of the images, decision making, object referencing, morphological feature extraction, arithmetic operations, and statistical analysis. This invention provides significantly faster phagocytic functions analysis than manual microscopic examination and more detailed quantitative morphological data than flow cytometery.

57 citations


Journal ArticleDOI
TL;DR: A new method is introduced for describing signature images that involves features of the signatures and relations among its parts organized in a hierarchical structure to classify and globally analyse their structure.

53 citations


Journal ArticleDOI
TL;DR: A technique for automatically clustering linear envelopes of EMGs (electromyograms) during gait has been developed that uses a temporal feature representation and a maximum peak matching scheme.
Abstract: A technique for automatically clustering linear envelopes of EMGs (electromyograms) during gait has been developed. It uses a temporal feature representation and a maximum peak matching scheme. This technique provides a viable way to define compact and meaningful EMG waveform features. The envelope matching is performed by dynamic programming, providing qualitatively the largest number of matched peaks and quantitatively a minimum distance measurement. The resulting averaged EMG profiles have low statistical variation and can serve as templates for EMG comparison and further classification. >

Proceedings ArticleDOI
03 Apr 1990
TL;DR: A texture analysis approach superior to previous ones in such aspects as classification/segmentation performance and applicability is presented, based on a widely adopted human visual model which hypothesizes that the HVS processes input pictorial signals through a set of parallel and quasi-independent mechanisms or channels.
Abstract: A texture analysis approach superior to previous ones in such aspects as classification/segmentation performance and applicability is presented. It is based on a widely adopted human visual model which hypothesizes that the human visual system (HVS) processes input pictorial signals through a set of parallel and quasi-independent mechanisms or channels. This model is referred to as the multichannel spatial filtering model (MSFM). The core of the MSFM presently applied is the recently formulated cortical channel model (CCM), which attempts to model the process of texture feature extraction in each individual channel in the MSFM. With these models, successful algorithms for both texture classification and segmentation (texture edge detection) have been developed. The algorithm for texture feature extraction and classification is compared with the conventional benchmark, i.e., the gray-level cooccurrence matrix approach, and proves to be superior in many aspects. The algorithm for texture edge detection is tested under a variety of textured images, and good segmentation results are obtained. >

Proceedings ArticleDOI
01 Sep 1990
TL;DR: The system can get the facial features precisely, automatically and independent of facial image size and face tilting using information about color and position of face and face components, and image histogram and line segment analysis.
Abstract: We have studied a stereo-based approach to three-dimensional face modeling and facial image reconstruction virtually viewed from different angles. This paper describes the system, especially image analysis and facial shape feature extraction techniques using information about color and position of face and face components, and image histogram and line segment analysis. Using these techniques, the system can get the facial features precisely, automatically and independent of facial image size and face tilting. In our system, input images viewed from the front and side of the face are processed as follows: the input images axe first transformed into a set of color pictures with significant features. Regions are segmented by thresholding or slicing after analyzing the histograms of the pictures. Using knowledge about color and positions of the face, face and hair regions are obtained and facial boundaries extracted. Feature points along the obtained profile are extracted using information about curvature amplitude and sign, and knowledge about distance between the feature points. In the facial areas which include facial components, regions are again segmented by the same techniques with color information from each face component. The component regions are recognized using knowledge of facial component position. In each region, the pictures are filtered with various differential operators, which are selected according to each picture and region. Thinned images are obtained from the filtered images by various image processing and line segment analysis techniques. Then, feature points of the front and side views are extracted. Finally, the size and position differences and facial tilting between two input images are compensated for by matching the common feature points in the two views. Thus, the three-dimensional data of the feature points and the boundaries of the face are acquired. The two base face models, representing a typical Japanese man and woman, are prepared and the model of the same sex is modified with 3D data from the extracted feature points and boundaries in a linear manner. The images, which are virtually viewed from different angles, are reconstructed by mapping facial texture to the modified model.

Journal ArticleDOI
01 Jul 1990
TL;DR: The performance of the system under different recognition-rejection tradeoff ratios is analyzed in detail and encouraging results on nearly 17000 totally unconstrained handwritten numerals are presented.
Abstract: A method of recognizing unconstrained handwritten numerals using a knowledge base is proposed. Features are collected from a training set and stored in a knowledge base that is used in the recognition stage. Recognition is accomplished by either an inference process or a structural method. The scheme is general, flexible, and applicable to different methods of feature extraction and recognition. By changing the acceptance parameters, a continuous range of performance can be achieved. Encouraging results on nearly 17000 totally unconstrained handwritten numerals are presented. The performance of the system under different recognition-rejection tradeoff ratios is analyzed in detail. >

Proceedings ArticleDOI
24 Sep 1990
TL;DR: The usefulness of Gabor filters in textural image processing and neural networks in image feature extraction is demonstrated and the importance of having good representations of image data to neural networks is illustrated through variations in performance of different trained networks.
Abstract: A system for minutiae extraction in fingerprint images using back-propagation networks and Gabor filters is described. Fingerprint images are first convolved with complex Gabor filters and the resulting phase and magnitude signals are passed to networks to identify minutia regions. Promising results are obtained with good detection ratio and low false alarm rate. The importance of having good representations of image data to neural networks is illustrated through variations in performance of different trained networks. The usefulness of Gabor filters in textural image processing and neural networks in image feature extraction is demonstrated. >

Journal ArticleDOI
TL;DR: Maximum a posteriori (MAP) estimation is used, together with statistical models for the speckle noise and for the curve-generation process, to find the most probable estimate of the feature, given the image data.
Abstract: A method for finding curves in digital images with speckle noise is described. The solution method differs from standard linear convolutions followed by thresholds in that it explicitly allows curvature in the features. Maximum a posteriori (MAP) estimation is used, together with statistical models for the speckle noise and for the curve-generation process, to find the most probable estimate of the feature, given the image data. The estimation process is first described in general terms. Then, incorporation of the specific neighborhood system and a multiplicative noise model for speckle allows derivation of the solution, using dynamic programming, of the estimation problem. The detection of curvilinear features is considered separately. The detection results allow the determination of the minimal size of detectable feature. Finally, the estimation of linear features, followed by a detection step, is shown for computer-simulated images and for a SAR image of sea ice.

Proceedings ArticleDOI
13 May 1990
TL;DR: An algorithm is proposed for pose estimation based on the volume measurement of tetrahedra composed of feature-point triplets extracted from an arbitrary quadrangular target and the lens center of the vision system that makes it a potential candidate for real-time robotic tasks.
Abstract: An algorithm is proposed for pose estimation based on the volume measurement of tetrahedra composed of feature-point triplets extracted from an arbitrary quadrangular target and the lens center of the vision system. Using a pinhole model (lens distortion is taken into account separately) and a quadrangular target, for which only the six distance measurements between all pairs of feature points are known, the complete pose is determined using an all-geometric closed-form solution for the six parameters of the pose (three translation components and three rotation components). This method has been tested using synthetic and real data and shown to be efficient, accurate, and robust. Its speed, in particular, makes it a potential candidate for real-time robotic tasks. >

Journal ArticleDOI
TL;DR: This paper develops a group theoretical model for the feature extraction part of pattern recognition systems and shows why the so found filter functions often appear as solutions to optimality problems and why they often have some nice properties such as invariance under Fourier transformation.

Book ChapterDOI
01 Jan 1990
TL;DR: In the context of computer vision, the recognition of three-dimensional objects typically consists of image capture, feature extraction, and object model matching.
Abstract: In the context of computer vision, the recognition of three-dimensional objects typically consists of image capture, feature extraction, and object model matching. During the image capture phase, a camera senses the brightness at regularly spaced points, or pixels, in the image. The brightness at these points is quantized into discrete values; the two-dimensional array of quantized values forms a digital image, the input to the computer vision system. During the feature extraction phase, various algorithms are applied to the digital image to extract salient features such as lines, curves, or regions. The set of these features, represented by a data structure, is then compared to the database of object model data structures in an attempt to identify the object. Clearly, the type of features that need to be extracted from the image depends on the representation of objects in the database.

Proceedings Article
01 Jan 1990
TL;DR: An overview of techniques for document image analysis can be found in this article, with an emphasis on those for grnphics recognition and interpretation, which is derived from the fields of image processing pattern recognition, and machine vision.
Abstract: An overview is presented of algorithms and techniques for document image analysis with an emphasis on those for grnphics recognition and interpretation The techniques are derived from the fields of image processing pattern recognition, and machine vision The objective in document image analysis is to recognize page contents including layout, text, and figures Although optical character recognition (OCR) fds within the context of document image analysis we do not cover this area since OCR techniques have been covered extensively in the literature We also limit the focus to images containing binary information Topics covered are segmentation of document image into text and graphics regions, vectorization to obtain lines, identification of graphical primitives, and generation of succinct image interpretations

Proceedings ArticleDOI
16 Jun 1990
TL;DR: A parallel implementation of a system to recognize 2D objects under realistic scenarios (occlusion, rotation, translation, and perspective) is presented, achieving O(n/sup -x/) with n/Sup x/ processors (x
Abstract: A parallel implementation of a system to recognize 2D objects under realistic scenarios (occlusion, rotation, translation, and perspective) is presented. A preprocessing phase and a recognition phase are used. Both phases have been implemented on the Connection Machine, achieving O(n/sup -x/) with n/sup x/ processors (x >

Proceedings ArticleDOI
01 Feb 1990
TL;DR: In this paper, an image sensor is set at an intersection to measure traffic flow by vehicle tracking, which is done by gray level pattern matching, and then the hardware for high speed gray-level pattern matching is provided.
Abstract: In this paper, we show anew image sensor for measuring traffic flow. The image sensor is set at an intersection. The image sensor measures traffic flow by vehicle tracking. The tracking is done by gray level pattern matching. Gray level pattern matching is suitable for outdoor use since it is robust against change of scene brightness. Then we provide the hardware for high speed gray level pattern matching. After showing our new method for measuring traffic flow by the image sensor, we measure traff~c flow at an actual intersection and describe the results. The sensor measures traffic flow accurately under various conditions.

Proceedings ArticleDOI
16 Jun 1990
TL;DR: The finger image identification method described uses the image of the entire finger, especially the finger joint line patterns, instead of the fingerprint, which makes it possible to verify a person's identity in less than 0.1 s.
Abstract: The finger image identification method described uses the image of the entire finger, especially the finger joint line patterns, instead of the fingerprint. This method makes it possible to verify a person's identity in less than 0.1 s. Since the feature is extracted by integrating a finger image, it is not sensitive to noise. Because the algorithm is simple, the software/hardware system can be realized at a low cost. Experiments based on the finger features of 508 people resulted in an error rate of 9.6% type-1 errors and 0.1% type-2 errors. >

Journal ArticleDOI
TL;DR: In this paper, an auto-regressive model and the corresponding processes of feature extraction and condition monitoring are investigated in detail, and an orthogonal transformation is introduced for both feature compression and the establishment of a statistical model of the template representing the normal condition.

Proceedings ArticleDOI
17 Jun 1990
TL;DR: The author begins by considering the decomposition of a feedforward network into a feature-discovery and a supervised-learning modules, then he introduces supervised feature discovery, and finally he describes a modular backpropagation algorithm.
Abstract: A modular architecture for supervised learning is presented. Three ways to modularize the learning are investigated: (1) preprocessing the input by a self-organizing subnetwork to extract strong features from the data, (2) supervised feature extraction, and (3) an iterative learning cycle in which only one layer learns at a time, with the output-layer weights learned by an exact method. With this modular architecture, only a small fraction of connection weights is determined by the gradient-descent method. A series of computational experiments shows the superiority of the modular model in learning quality and speed. The author begins by considering the decomposition of a feedforward network into a feature-discovery and a supervised-learning modules, then he introduces supervised feature discovery, and finally he describes a modular backpropagation algorithm

Journal ArticleDOI
TL;DR: A new scheme using a pair of stereo images for 3-D information derivation about a scene is presented and a triangulation geometry between two camera models and a point in3-D scene is established.

Proceedings ArticleDOI
03 Apr 1990
TL;DR: Experimental results comparing the performances of the multiple-codebook and single- codebook methods indicate that, when the codebook size is small, the multiple's method is better than the single's method, however, if the codebooks size is reasonably large, the single-codebooks method displays better performance than the multiple’s method.
Abstract: A vector quantization (VQ)-based recognition method which uses feature vector codebooks containing hierarchical spectral dynamics is proposed. This method is highly effective for reducing the number of candidates in word recognition and achieving a high recognition accuracy in /b/,/d/ and /g/ recognition. Since this method does not need time alignment, it has the advantage of a small amount of computation and ease of parallel processing. Experimental results comparing the performances of the multiple-codebook and single-codebook methods indicate that, when the codebook size is small, the multiple-codebook method is better than the single-codebook method. However, if the codebook size is reasonably large, the single-codebook method displays better performance than the multiple-codebook method. >

Proceedings ArticleDOI
16 Jun 1990
TL;DR: The authors describe the geometrical criteria which define viewpoint-invariant features to be extracted from 2-D line drawings of 3-D objects and discuss the extraction of these features, which forms the initial stage of a generic object recognition system, the Primal Access Recognition of Visual Objects (PARVO) system.
Abstract: The authors describe the geometrical criteria which define viewpoint-invariant features to be extracted from 2-D line drawings of 3-D objects. They also discuss the extraction of these features, which forms the initial stage of a generic object recognition system, the Primal Access Recognition of Visual Objects (PARVO) system. In this system, part-based qualitative descriptions are built and matched to coarse 3-D object models for recognition. The segmentation and labeling of the constituent parts of an object rely on the 3-D properties inferred from the presence of its 2-D features. The original motivation for PARVO its recognition by components, a theory of human image understanding from the field of psychology. Definitions of the geometrical criteria defining the viewpoint-invariant features are introduced. Examples of results obtained by applying these criteria to a typical line drawing are shown. >

Proceedings ArticleDOI
16 Jun 1990
TL;DR: An enhanced border-following algorithm and its application to document image processing is presented and various kinds of components in a document image can be flexibly segmented and extracted with a variable-size mask for border following instead of the conventional 3*3-size Mask.
Abstract: An enhanced border-following algorithm and its application to document image processing is presented. Various kinds of components (characters, text lines, text blocks, figures, tables, etc.) in a document image can be flexibly segmented and extracted with a variable-size mask for border following instead of the conventional 3*3-size mask. An automatic document image structuring process to construct a multimedia document and a raster/geometric conversion method for the segmented graphic parts of the image. such as diagrams and tables, are discussed. >

Patent
30 Aug 1990
TL;DR: In this paper, a primary processing device is used to extract the characteristic feature of an object pattern from an original image, and the information as a result of the above processing is inputted into the input of the all-optical type optical neural network apparatus, that implements parallel processings adaptively through optical computing at individual points on the input.
Abstract: For inputting a two-dimensional image into an optical neural network apparatus, a primary processing device is used to extract the characteristic feature of an object pattern. Thereafter, compressed information as a result of the above processing is inputted into the input of the all-optical type optical neural network apparatus, that implements parallel processings adaptively through optical computing, at individual points on the input of the same. Therefore, the primary processing device that was capable of dealing with only logical input information until now can process even vague input information by the use of the optical neural network apparatus located on the later stage. On the other hand, the use of the primary processing device on the previous stage of the optical neural network apparatus enables a limited input range of the optical neural network apparatus to be expanded together with the assurance of higher degree processing by inputting into the optical neural network apparatus results of the characteristic feature extraction from an original image.