Scheduled System Maintenance:
On May 6th, single article purchases and IEEE account management will be unavailable from 8:00 AM - 12:00 PM ET (12:00 - 16:00 UTC). We apologize for the inconvenience.
By Topic

Applications of Neural Networks to Signal Processing (Digest No. 1994/248), IEE Colloquium on

Date 15 Dec 1994

Filter Results

Displaying Results 1 - 12 of 12
  • Gesture recognition: an assessment of the performance of recurrent neural networks versus competing techniques

    Publication Year: 1994 , Page(s): 8/1 - 8/3
    Cited by:  Papers (1)
    Save to Project icon | Click to expandQuick Abstract | PDF file iconPDF (212 KB)  

    A gesture is a motion of the body that contains information (e.g. waving goodbye, beckoning with an index finger, signs in a sign language). There are four classes of gestures; signs (substitutes for spoken language); indications (pointing and showing direction); illustration (conveying ideas such as size and shape); and manipulation (for example making something from virtual clay). The first three of these are suitable for both input and output, while the fourth is only suitable for input. Recognition of gestures is still a major problem, and represents a challenge that rivals speech and hand-writing recognition. The paper describes a comparison of some of the competing techniques that have been applied to solving this problem. Three techniques were investigated; dynamic programming (DP), hidden Markov models (HMMs) and recurrent neural networks (RNNs). All of these techniques seek to represent time explicitly, and are therefore better suited than static techniques to the dynamic nature of most gestures. The study has application to a sign language recognition system View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Techniques for detection and classification of the fetal QRS complex

    Publication Year: 1994 , Page(s): 10/1 - 10/3
    Save to Project icon | Click to expandQuick Abstract | PDF file iconPDF (216 KB)  

    A critical process in both the fetal cardiotocogram (CTG) and fetal electrocardiogram (FECG) analysis is to determine the location of the QRS complex from the raw FECG data. Difficulties arise because the FECG signal is degraded and sometimes totally obscured by noise. Thus, to determine the location of the QRS complexes is a two stage process: pre-processing of the noisy FECG to detect candidate QRS complexes, and pattern recognition to classify these as genuine QRS complexes or noise. The authors compare an ANN technique with conventional techniques for both pre-processing and pattern recognition of the FECG View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Comparing classical and neural network classification techniques for image feature identification

    Publication Year: 1994 , Page(s): 1/1 - 1/8
    Save to Project icon | Click to expandQuick Abstract | PDF file iconPDF (476 KB)  

    One of the main requirements of an image processing system is the ability to automatically recognise a given object within a scene. Many military systems rely on the use of imagery based upon infra-red (IR) technology. Another requirement is for robustness over a wide range of operating conditions. Another overall consideration in any system is one of processing requirements in terms of speed, cost and physical size; military systems often impose severe constraints on the physical size. A simple approach to recognising objects is to segment the image to form a set of regions, where one or more of the regions is representative of the object. This step is then followed by the computation of a set of rudimentary numerical metrics for each object region indicated by the segmentation process. For each region, the set of metrics form a feature vector whose values are, in some sense, representative of the nature of the region. Ideally the elements of the feature vector associated with a object region of one class will significantly differ from the feature vectors describing regions from other classes. So the objective is to classify the object region given the contents of the feature vector computed for that region. Traditionally, a typical feature classification process made use of well established algorithms based upon linear discriminant or nearest neighbour techniques. More recently neural network classification techniques have emerged and the objective of the present paper is to perform some initial simple experiments to compare the traditional and more contemporary classification techniques View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • IEE Colloquium on `Applications of Neural Networks to Signal Processing' (Digest No.1994/248)

    Publication Year: 1994
    Save to Project icon | Click to expandQuick Abstract | PDF file iconPDF (48 KB)  

    The following topics were dealt with: image feature identification; edge detection; image noise suppression; remote serving data classification; texture classification; ESM radar function identification; gesture recognition; medical signal processing; adaptive equalisers in digital communications View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Neural network edge detection-successes and failures

    Publication Year: 1994 , Page(s): 2/1 - 2/3
    Cited by:  Papers (1)
    Save to Project icon | Click to expandQuick Abstract | PDF file iconPDF (216 KB)  

    A common problem in image processing is the detection of edges in noisy and incomplete images. Conventional edge detection techniques rely on local gradients which are not robust in noise. Variable thresholding can be used to detect changing edge strengths in the image, but these thresholds have to be found. The present work examines the use of various neural network topologies to improve the robustness of the edge detection in noisy and incomplete images. Throughout the paper neural network edge detection is illustrated by the left ventricle boundary extraction problem in echocardiographic images View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Comparing small object detectors

    Publication Year: 1994 , Page(s): 4/1 - 4/3
    Cited by:  Papers (1)
    Save to Project icon | Click to expandQuick Abstract | PDF file iconPDF (188 KB)  

    The signal produced by a small object when imaged on a CCD array is spread over several pixels in the array due to device physics and atmospheric effects. It has been typically represented as a raised cosine function. The present paper details a comparison of two potential small object detection networks: (a) the artificial neural network (ANN) and, (b) the least-mean-square (LMS) filter View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • How do neural networks compare with standard filters for image noise suppression?

    Publication Year: 1994 , Page(s): 3/1 - 3/4
    Save to Project icon | Click to expandQuick Abstract | PDF file iconPDF (444 KB)  

    The present paper has been partly motivated by curiosity-can ANNs successfully cope with image noise removal? If so, can they improve on recognised noise suppression techniques? One must remember that the latter use conventional algorithms, or the corresponding hardware, and are not trained to perform the task. Yet the very fact of training reflects that ANNs learn by example, embodying implicit learning rules-thereby emulating biological systems and providing the potential to improve on conventional algorithmic approaches. In fact, there are possibilities that ANNs might perform noise suppression more effectively than conventional approaches, not least in adapting to specific types of noise, and in eliminating the image distortion which is a characteristic of the widely used median filter. In this context it is worth noting that the median filter has no adjustable parameters other than neighbourhood size, so ANNs definitely have the potential for improving on its performance-and also on that of alternative types of filter. The paper describes the authors' own studies of the problem View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Classification of multi-spectral remote sensing data with neural networks: a comparative study

    Publication Year: 1994 , Page(s): 5/1 - 5/2
    Save to Project icon | Click to expandQuick Abstract | PDF file iconPDF (148 KB)  

    Satellites or planes generate remote sensing images by simultaneously recording `grey-level' images for a number of wave-bands. The resulting images are usually processed using statistical classifiers to extract features such as roads, built-up areas, vegetation or water. In the present study two types of neural networks, a multi-layer perceptron (MLP) and a Kohonen learning vector quantization (LVQ) network are tested as pattern classifiers. The results are compared with a nearest neighbour classifier (KNN). The aim of the study is to extract five classes: (1) roads, (2) buildings, (3) vegetation, (4) water and (5) derelict sites from data obtained using multi-spectral images of Stoke-on-Trent with a pixel resolution of roughly 4×4 m. The architecture and learning parameters of each network were optimised for 4005 training pixels selected randomly over the image (891×3989 pixels). Both network types and the statistical classifier were tested on 3552 test patterns. Standard back-propagation was used to train the MLPs while oLVQ1 and LVQ3 training were used for the LVQ networks View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A critical assessment of recurrent artificial neural networks as adaptive equalizers in digital communications

    Publication Year: 1994 , Page(s): 11/1 - 11/4
    Cited by:  Papers (1)
    Save to Project icon | Click to expandQuick Abstract | PDF file iconPDF (208 KB)  

    A number of neural network structures have previously been applied to the problem of equalization of digital communications channels and view the problem as one of pattern classification rather than one of inverse filtering. The recurrent neural network (RNN) has previously been shown to outperform the conventional linear transversal equalizer structure and has the advantage of requiring a small number of nodes to achieve a given level of equalization. The paper aims to highlight the mechanism by which RNNs equalize channels and to show that the dynamics of such networks create a structure unsuitable for reliable equalization View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • The application of artificial neural networks to naval ESM radar function identification

    Publication Year: 1994 , Page(s): 7/1 - 7/3
    Save to Project icon | Click to expandQuick Abstract | PDF file iconPDF (236 KB)  

    Identification is the last stage in ESM (electronic surveillance measures) processing, which is currently the radar type classification of each track. A track is a parametric description of a radar signal thought to be in the environment that results from the preceding stages of pulse sorting/grouping and characterisation. Identification consists essentially of an ESM radar library and an identification algorithm. The library describes the characteristics of known radars to be used by the identification algorithm. The identification algorithm defines the method by which the parameters of the track are matched against the library information and the result is a radar type identity for the track View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Neural networks and texture classification

    Publication Year: 1994 , Page(s): 6/1 - 6/4
    Save to Project icon | Click to expandQuick Abstract | PDF file iconPDF (232 KB)  

    Texture plays an increasingly important role in computer vision. It has found wide application in remote sensing, medical diagnosis, quality control, food inspection and so forth. Research on texture started in the 1970s. The resurgence of research interest and resulting techniques in artificial neural networks gives rise to a new paradigm for texture analysis. The paper presents an application of a neural network architecture along with its training algorithm-the generating-shrinking algorithm-to texture classification in comparison with the error backpropagation algorithm and the conventional K-nearest neighbour rule. The texture feature sets considered in the paper include the statistical geometrical features and features derived from the two-dimensional discrete Fourier transform via rings and wedges View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Application of artificial neural networks to medical signal processing

    Publication Year: 1994 , Page(s): 9/1 - 9/3
    Save to Project icon | Click to expandQuick Abstract | PDF file iconPDF (228 KB)  

    The dynamics of human sleep have previously been examined using unsupervised clustering techniques [Roberts and Tarassenko, 1992]. This culminated in the hypothesis that the structure of sleep can be described as a linear combination of three underlying processes. These correspond to the conventional, rule-based stages of wakefulness, REM sleep, and the deepest form of non-REM sleep, stage 4. The mixing fractions, p(W), p(R), and p(S), of these three processes vary as sleep progresses, and to estimate them a system has been developed that comprises an autoregressive (AR) model [Makhoul, 1975, Kay and Marple, 1981] followed by two artificial neural networks: a multi-layer perceptron (MLP) and a radial basis function (RBF) network, operating in parallel. The AR model is used to pre-process the EEG on a second-by-second basis, while the mixing fractions for each second are then estimated using the neural networks. The system is currently undergoing clinical trials, during which time the performance of the MLP and RBF networks will be assessed and a choice made as to which one to retain in the final, commercial system View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.