Showing papers by "Anthony G. Constantinides published in 1995"

LMS+F algorithm

Abstract: A stochastic gradient adaptive filter algorithm using a time-varying mixed criterion is proposed. The algorithm optimises a cost function based on an adaptively adjusted combination of the LMF and LMS functions. Two different combination strategies are studied and sufficient convergence conditions are presented. The time-varying nature of the cost function adds flexibility and enables rapid response of the algorithm.

Finite-precision design and implementation of all-pass polyphase networks for echo cancellation in sub-bands

Abstract: All-pass polyphase networks (APN) are particularly attractive for acoustical echo cancellation (AEC) arranged in sub-bands. They provide lower inter-band aliasing, delay and computational complexity than their FIR counterparts. Moreover, APNs achieve higher echo return loss enhancement (ERLE) performance and faster convergence than full-band processing. In the paper, the finite precision implementation of APNs is addressed. A procedure is presented for re-optimising the all-pass coefficients of the prototype low-pass filter for finite precision operation. Robust finite precision implementation of a prototype low-pass filter is discussed. The results of a set of AEC experiments are reported with full and 16-bit precision implementation.

Object oriented motion and deformation estimation using composite segmentation

Abstract: A novel object oriented motion estimation algorithm is presented. The algorithm provides the means for highly efficient moving image encoding by fully exploiting the temporal redundancy among the objects of successive frames. Two-dimensional segmentation is performed on a composite image synthesised from two consecutive frames. The object correspondence problem is removed implicitly by virtue of the fact that the generated composite segments correspond to successive versions of the same objects. The scheme guarantees well matched segments by reducing the effects of noise and varying illumination. While preserving motion or deformation information. Progressive motion estimation is achieved within the segmentation process which adapts to the assumed translational or affine model. Motion compensated extrapolation is performed on uncovered background and overlapping regions of the predicted frame. Simulation results show clearly the efficiency of the predictive scheme even in the case when only motion and deformation parameters need to be transmitted.

Root moments: an alternative interpretation of cepstra for signal feature extraction and modelling

Abstract: A new approach for signal modelling that depends on the algebraic properties of the signal under consideration has been developed. It renders many important signal processing problems easily tractable. This article presents the elements of the theory behind such algebraic modelling. The fundamental concepts behind the approach rely on the root-moment properties of polynomials which have been first formulated by Newton and known as the Newton identities.

Lagrange programming neural networks for blind Volterra system modelling

Abstract: In this paper the problem of nonlinear signal modelling is examined from a mixed order statistical perspective. The approach taken involves the use of second order Volterra kernels which are derived from a joint operation on second and third order moments of the signals. The paper describes the fundamental issues of the various components of the approach both for one dimensional and two dimensional signals. The nonlinear equations involved are solved by means of unconstrained Lagrange programming neural networks. The Volterra kernels may be used in further operations such as features for image classification and segmentation.

Object oriented motion and deformation estimation using composite image segmentation

Abstract: A novel object oriented motion estimation algorithm is presented. The algorithm provides the means for highly efficient moving image encoding by fully exploiting the temporal redundancy among the objects of successive frames. Two-dimensional segmentation is performed on a composite image synthesised from two consecutive frames. The object correspondence problem is removed implicitly by virtue of the fact that the generated composite segments correspond to successive versions of the same objects. Thus the scheme not only solves the problem of the correspondence between successive versions of the same object, but it also guarantees well matched segments in the presence of noise and varying illumination. Moreover, it preserves motion or deformation information. Progressive motion estimation is achieved within the segmentation process which adapts to the assumed translational or affine model. Motion compensated extrapolation is performed on uncovered background and overlapping regions of the predicted frame. Simulation results prove the efficiency of the predictive scheme even in the case that only motion and deformation parameters need to be transmitted. >

Problem-independent parallel realisation of simulated annealing on a ring multiprocessor architecture based on speculative computation

Abstract: In this paper, a novel problem-independent, parallel realisation of the Simulated Annealing (SA) algorithm is proposed. By employing speculative computation, concurrency is introduced into the inherently sequential algorithm. This is achieved by predicting the acceptance of each generated move before the move is evaluated. Based on the prediction, subsequent moves can be proposed and evaluated before decisions on whether to accept or reject preceding moves are made. To preserve the sequential decision nature of SA. all moves subsequent to a prediction that is eventually proved wrong are discarded. A simple and effective prediction mechanism using previous move statistics is developed. Efficient realisation of the parallel SA algorithm on a ririg multiprocessor architecture is described. Analytical and simulation performance results are presented. These results indicate that our parallel SA is best implemented on a coarse to medium grain rnultiprocessor system