Konstantinos I. Diamantaras

ECG pattern recognition and classification using non-linear transformations and neural networks: A review

The most widely used signal in clinical practice is the ECG. ECG conveys information regarding the electrical function of the heart, by altering the shape of its constituent waves, namely the P, QRS, and T waves. Thus, the required tasks of ECG processing are the reliable recognition of these waves, and the accurate measurement of clinically important parameters measured from the temporal distribution of the ECG constituent waves. In this paper, we shall review some current trends on ECG pattern recognition. In particular, we shall review non-linear transformations of the ECG, the use of principal component analysis (linear and non-linear), ways to map the transformed data into n-dimensional spaces, and the use of neural networks (NN) based techniques for ECG pattern recognition and classification. The problems we shall deal with are the QRS/PVC recognition and classification, the recognition of ischemic beats and episodes, and the detection of atrial fibrillation. Finally, a generalised approach to the classification problems in n-dimensional spaces will be presented using among others NN, radial basis function networks (RBFN) and non-linear principal component analysis (NLPCA) techniques. The performance measures of the sensitivity and specificity of these algorithms will also be presented using as training and testing data sets from the MIT-BIH and the European ST-T databases.

Adaptive Principal component EXtraction (APEX) and applications

The authors describe a neural network model (APEX) for multiple principal component extraction. All the synaptic weights of the model are trained with the normalized Hebbian learning rule. The network structure features a hierarchical set of lateral connections among the output units which serve the purpose of weight orthogonalization. This structure also allows the size of the model to grow or shrink without need for retraining the old units. The exponential convergence of the network is formally proved while there is significant performance improvement over previous methods. By establishing an important connection with the recursive least squares algorithm they have been able to provide the optimal size for the learning step-size parameter which leads to a significant improvement in the convergence speed. This is in contrast with previous neural PCA models which lack such numerical advantages. The APEX algorithm is also parallelizable allowing the concurrent extraction of multiple principal components. Furthermore, APEX is shown to be applicable to the constrained PCA problem where the signal variance is maximized under external orthogonality constraints. They then study various principal component analysis (PCA) applications that might benefit from the adaptive solution offered by APEX. In particular they discuss applications in spectral estimation, signal detection and image compression and filtering, while other application domains are also briefly outlined. >

Blind two-input-two-output FIR channel identification based on frequency domain second-order statistics

We present an analytical solution to the two-input-two-output blind crosswise mixture identification based on eigenvalue decomposition of second-order spectra correlations. The sources are independent and non-white, but otherwise, we consider their statistics to be unknown. We show that the cross channels cannot be uniquely determined by the analysis of the frequency domain covariance alone due to the unknown eigenvector permutations. However, the problem can be attacked with the help of two invariant indices that are immune to these permutations. Using these indices together with standard reconstruction-from-phase techniques, we show that the channels can be uniquely determined. Our theoretical results lead to a novel frequency domain second-order algorithm that identifies the unknown channels.

Blind MIMO FIR channel identification based on second-order spectra correlations

We consider a problem of identifying a multiple-input multiple-output (MIMO) finite impulse response (FIR) system excited by colored inputs with known statistics. We propose a new, nonlinear optimization-based method that involves the power spectra and cross-spectra of the system output. The proposed algorithm is tested for the case of cyclostationary inputs (CDMA scenario) and stationary inputs (SDMA scenario). Simulation results indicate that the proposed scheme works well, even for large order systems, and is robust to noise and channel length mismatch.

Blind Deconvolution of Multi-Input Single-Output Systems With Binary Sources

The problem of blind source separation for multi-input single-output (MISO) systems with binary inputs is treated in this paper. Our approach exploits the constellation properties of the successor values for each output sample. In the absence of noise, the successors of each output value form a characteristic finite set of clusters (successor constellation). The shape of this constellation is invariant of the predecessor value and it only depends on the last filter tap. Consequently, the localization of the successors constellation can lead to the removal of the last filter tap, thus reducing the length of the filter-a process we call channel deflation. Based on the successor observation clustering (SOC), we develop two algorithms for blind source separation-SOC-1 and SOC-2-differing mainly on the required size of the data set. Furthermore, the treatment of the system in the presence of noise is described using data clustering and data correction. The problem of noise is attacked using a statistical-mode-based method. Moreover, we correct the problem of misclassified observations using an iterative scheme based on the Viterbi algorithm for the decoding of a hidden Markov model (HMM)

Blind separation of multiple binary sources using a single linear mixture

Multiple binary sequences are blindly separated from a single linear mixture using the structure of the probability distribution function of the observed data. No specific assumptions are made regarding the second or higher order statistics of the sources. Assuming additive Gaussian noise the PDF is a mixture of Gaussians centered at points that uniquely determine (a) the mixing parameters and (b) the source signals up to a permutation and a sign ambiguity. We present both the theoretical framework for this novel approach and a new recursive blind separation algorithm based on our framework. Simulations show that the method can successfully separate at least up to 10 binary source signals at different noise levels.

A Linear Systolic Array for Real-Time Morphological Image Processing

Mathematical morphology has proven to be a very useful tool for applications such as smoothing, image skeletonization, pattern recognition, machine vision, etc. In this paper we present a 1-dimensional systolic architecture for the basic gray-scale morphology operations: dilation and erosion. Most other morphological operations like opening and closing, are also supported by the architecture since these operations are combinations of the basic ones. The advantages of our design stem from the fact that it has pipeline period α = 1 (i.e., 100% processor utilization), it requires simple communications, and it is exploiting the simplicity of the morphological operations to make it possible to implement them in a linear target machine although the starting algorithm is a generalized 2-D convolution. We also propose a Locally Parallel Globally Sequential (LPGS) partitioning strategy for the best mapping of the algorithm onto the architecture. We conclude that for this particular problem LPGS is better than LSGP in a practical sense (pinout, memory requirement, etc.). Furthermore, we propose a chip design for the basic component of the array that will allow real-time video processing for 8- and 16-bit gray-level frames of size 512 × 512, using only 32 processors in parallel. The design is easily scalable so it can be custom-taylored to fit the requirement of each particular application.

Occlusion resistant object tracking

Object tracking with occlusion prediction using multiple feature correspondences is proposed. The tracking region is defined by a set of point features, tracked using Kanade-Lucas-Tomasi (1991) algorithm. During total occlusion the region position is estimated using motion prediction based on a Kalman filtering scheme applied to the motion model prior to occlusion. During partial occlusion the displacements of the occluded features are predicted based on the motion of the bounding box of the moving object. Experimental results on real and artificial images have shown that the algorithm behaves well under total and partial occlusion.

Robust estimation of rigid-body 3-D motion parameters based on point correspondences

The estimation of rigid-body 3-D motion parameters using point correspondences from a pair of images under perspective projection is, typically, very sensitive to noise. We present a novel robust method combining two approaches: (1) the SVD analysis of a linear operator resulting from the feature points and the displacement vectors and (2) a modified version of the well-known weighted least-squares method proposed by Huber in the context of robust statistics. We give a detailed rank analysis of the involved linear operator and study the effects of noise. We also propose a robust method guided by the structure of this operator, using weighted least squares and data partitioning. The method has been tested on artificial data and on real image sequences showing a remarkable robustness, even in the presence of up to 50% outliers in the data set.

An Efficient Subspace Method for the Blind Identification of Multichannel FIR Systems

We present a novel method for the blind identification of linear, single-input multiple-output (SIMO) finite- impulse-response (FIR) systems, based on second-order statistics. Our approach, called the truncated transfer matrix method (TTM) proceeds in two major steps: first, the SVD analysis of the lagged covariance matrix gives the subspace of the clipped system transfer matrix and second, the block-Toeplitz structure of the transfer matrix gives extra constraints that allow us to reconstruct the matrix through the solution of a linear system of equations. The proposed TTM method is analytical (no optimization procedure involved), and it is robust to noise. We find that the method comes with an increased computational cost but it significantly outperforms state of the art second-order methods in low signal-to-noise ratio (SNR) situations.