By Topic

Vision, Image and Signal Processing, IEE Proceedings -

Issue 4 • Date Aug 2002

Filter Results

Displaying Results 1 - 9 of 9
  • Robustness evaluation of a minimal RBF neural network for nonlinear-data-storage-channel equalisation

    Publication Year: 2002 , Page(s): 211 - 216
    Save to Project icon | Click to expandQuick Abstract | PDF file iconPDF (403 KB)  

    The authors present a performance-robustness evaluation of the recently developed minimal resource allocation network (MRAN) for equalisation in highly nonlinear magnetic recording channels in disc storage systems. Unlike communication systems, equalisation of signals in these channels is a difficult problem, as they are corrupted by data-dependent noise and highly nonlinear distortions. Nair and Moon (1997) have proposed a maximum signal to distortion ratio (MSDR) equaliser for data storage channels, which uses a specially designed neural network, where all the parameters of the neural network are determined theoretically, based on the exact knowledge of the channel model parameters. In the present paper, the performance of the MSDR equaliser is compared with that of the MRAN equaliser using a magnetic recording channel model, under Conditions that include variations in partial erasure, jitter, width and noise power, as well as model mismatch. Results from the study indicate that the less complex MRAN equaliser gives consistently better performance robustness than the MSDR equaliser in terms of signal to distortion ratios (SDRs) View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Use of periodic and monotonic activation functions in multilayer feedforward neural networks trained by extended Kalman filter algorithm

    Publication Year: 2002 , Page(s): 217 - 224
    Cited by:  Papers (2)
    Save to Project icon | Click to expandQuick Abstract | PDF file iconPDF (611 KB)  

    The authors investigate the convergence and pruning performance of multilayer feedforward neural networks with different types of neuronal activation functions in solving various problems. Three types of activation functions are adopted in the network, namely, the traditional sigmoid function, the sinusoidal function and a periodic function that can be considered as a combination of the first two functions. To speed up the learning, as well as to reduce the network size, the extended Kalman filter (EKF) algorithm conjunct with a pruning method is used to train the network. The corresponding networks are applied to solve five typical problems, namely, 4-point XOR logic function, parity generation, handwritten digit recognition, piecewise linear function approximation and sunspot series prediction. Simulation results show that periodic activation functions perform better than monotonic ones in solving multicluster classification problems. Moreover, the combined periodic activation function is found to possess the fast convergence and multicluster classification capabilities of the sinusoidal activation function while keeping the robustness property of the sigmoid function required in the modelling of unknown systems View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Adaptive blind channel identification algorithm based on linear prediction for SIMO FIR systems

    Publication Year: 2002 , Page(s): 225 - 230
    Save to Project icon | Click to expandQuick Abstract | PDF file iconPDF (385 KB)  

    An adaptive algorithm for blind identification of single-input multiple-output (SIMO) FIR systems is proposed. It is based on the one-step forward linear prediction (LP) technique and can be implemented by an RLS adaptation. Unlike most second-order statistics (SOS)-based approaches, the proposed solution does not require the computation of the correlation matrix or its inverse explicitly. The obtained results demonstrate that the proposed approach is able to deliver better performance compared with the typical batch algorithm. It is also observed that the proposed algorithm can tolerate the appearance of near common zeros among the subchannels View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Taylor series based two-dimensional digital differentiators

    Publication Year: 2002 , Page(s): 231 - 236
    Cited by:  Papers (6)
    Save to Project icon | Click to expandQuick Abstract | PDF file iconPDF (488 KB)  

    A new type of Taylor series based 2-D finite difference approximation is presented, and it is shown that the coefficients of these approximations are not unique. Explicit formulas are presented for one of the possible sets of coefficients for an arbitrary order, by extending the previously presented 1-D approximations. These coefficients are implemented as maximally linear 2-D FIR digital differentiators, and their formulas are modified to narrow the inaccuracy regions on the resultant frequency responses, close to the Nyquist frequencies View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Parallel genetic algorithm based unsupervised scheme for extraction of power frequency signals in the steel industry

    Publication Year: 2002 , Page(s): 204 - 210
    Cited by:  Papers (1)
    Save to Project icon | Click to expandQuick Abstract | PDF file iconPDF (520 KB)  

    A steel industry has different types of loads, and so the incoming supply voltage of some units becomes distorted thus affecting those systems that depend on a distortionless supply. A novel unsupervised scheme named the recursive hybrid parallel genetic algorithm based line enhancer (RHPGABLE) scheme is proposed, to track the desired power frequency signal from the corrupted one. The RHPGABLE scheme is based on a proposed new crossover operator known as the generalised crossover (GC) operator. The delay and the filter coefficients are estimated recursively to yield optimal solutions. In the recursion of the proposed RHPGABLE algorithm, a parallel genetic algorithm (PGA) based on a coarse-grained approach is employed to estimate the delay, while the filter coefficients are estimated by PGA and a least mean squares (LMS) algorithm. RHPGABLE is an unsupervised scheme in the sense that no a priori knowledge of delay or filter coefficients and the associated training signal component is assumed to be available. The proposed scheme has been tested successfully on both synthetic data and data obtained from the Steel Melting Shop of Rourkela Steel Plant, Orissa, India View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Well-posed anisotropic diffusion for image denoising

    Publication Year: 2002 , Page(s): 244 - 252
    Cited by:  Papers (7)
    Save to Project icon | Click to expandQuick Abstract | PDF file iconPDF (698 KB)  

    A nonlinear iterative smoothing filter based on a second-order partial differential equation is introduced. It smooths out the image according to an anisotropic diffusion process. The approach is based on a smooth approximation of the total variation (TV) functional which overcomes the non-differentiability of the TV functional at the origin. In particular, the authors perform linear smoothing over smooth areas but selective smoothing over candidate edges. By relating the smoothing parameter to the time step, they arrive at a CFL condition which guarantees the causality of the discrete scheme. This allows the adoption of higher time discretisation steps, while ensuring the absence of artefacts deriving from the non-smooth behaviour of the TV functional at the origin. In particular, it is shown that the proposed approach avoids the typical staircase effects in smooth areas which occur in the standard time-marching TV scheme View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Steady-state mean-square error analysis of the cross-correlation and constant modulus algorithm in a MIMO convolutive system

    Publication Year: 2002 , Page(s): 196 - 203
    Cited by:  Papers (3)
    Save to Project icon | Click to expandQuick Abstract | PDF file iconPDF (531 KB)  

    The cross-correlation and constant modulus algorithm (CC-CMA) has been proven to be an effective approach in the problem of joint blind equalisation and source separation in a multi-input and multi-output system. In the paper, the steady-state mean-square error performance of CC-CMA in a noise-free environment is studied, and a new expression is derived based on the energy preservation approach of Mai and Sayed (2000). Simulation studies are undertaken to support the analysis View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • New voice over Internet protocol technique with hierarchical data security protection

    Publication Year: 2002 , Page(s): 237 - 243
    Cited by:  Papers (4)
    Save to Project icon | Click to expandQuick Abstract | PDF file iconPDF (507 KB)  

    The authors propose a voice over Internet protocol (VoIP) technique with a new hierarchical data security protection (HDSP) scheme. The proposed HDSP scheme can maintain the voice quality degraded from packet loss and preserve high data security. It performs both the data inter-leaving on the inter-frame of voice for achieving better error recovery of voices suffering from continuous packet loss, and the data encryption on the intra-frame of voice for achieving high data security, which are controlled by a random bit-string sequence generated from a chaotic system. To demonstrate the performance of the proposed HDSP scheme, we have successfully verified and analysed the proposed approach through software simulation and statistical measures on several test voices View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Multi-output regression using a locally regularised orthogonal least-squares algorithm

    Publication Year: 2002 , Page(s): 185 - 195
    Cited by:  Papers (3)
    Save to Project icon | Click to expandQuick Abstract | PDF file iconPDF (593 KB)  

    The paper considers data modelling using multi-output regression models. A locally regularised orthogonal least-squares (LROLS) algorithm is proposed for constructing sparse multi-output regression models that generalise well. By associating each regressor in the regression model with an individual regularisation parameter, the ability of the multi-output orthogonal least-squares (OLS) model selection to produce a parsimonious model with a good generalisation performance is greatly enhanced View full abstract»

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