This paper tests a SAR image coregistration method, developed to account for a joint rotation and range/azimuth shift effect in absence of zooming, on foliage penetrating (FOPEN) data. In particular, the method is referred as a constrained Least Squares (CLS) optimization method and, in its basic form, it sharply extracts all patches composing the entire image. Differently, in next developments it applies a detection stage to identify extended areas in the images where patches are then selected. Moreover, it also performs a refinement of the equations in the CLS problem through an iterative cancellation procedure. The performance of this enhanced version of the CLS are made on the challenging Carabas-II VHF-band FOPEN SAR data to demonstrate its effectiveness also in high-resolution SAR images.

This letter extends the constrained least-squares (CLS) optimization method developed to coregister multitemporal synthetic aperture radar (SAR) images affected by a joint rotation effect and range/azimuth shifts enforcing the absence of zooming effects. To take advantage of the structural information extracted from the scene, the method starts with a detection stage that identifies extended targets/areas in the images. The selected tie-points allow the CLS problem to be reformulated to find its (initial) solution based on a robust subset of image blocks. Then, the mean square error (MSE) of each equation evaluated from the initial solution allows to implement an iterative cancellation procedure to further skim the CLS equation set. The effectiveness of the proposed procedure is validated on real SAR data in comparison with the standard CLS.

In this paper, a Modified Capsule Neural Network (Mod-CapsNet) with a pooling layer but without the squash function is used for recognition of indoor home scenes which are represented in grayscale. This Mod-CapsNet produced an accuracy of 70% compared to the 17.2% accuracy produced by a standard CapsNet. Since there is a lack of larger datasets related to indoor home scenes, to obtain better accuracy with smaller datasets is also one of the important aims in the paper. The number of images used for training and testing is 20,000 and 5000 respectively, all of dimension 128X128. The analysis proves that in the indoor home scene recognition task the combination of the capsule without a squash function and with max-pooling layers works better than by using capsules with convolutional layers. Indoor home scenes are specifically focused towards analysing capsules performance on datasets whose images have similarities but are, nonetheless, quite different. For example, tables may be present in living rooms and dining rooms even though these are quite different rooms.

Novel technologies for EMG (Electromyogram) based hand gesture recognition have been investigated for many industrial applications. In this paper, a novel approach which is based on a specific designed spiking convolution neural network which is fed by a novel EMG signal energy density map is presented. The experimental results indicate that the new approach not only rapidly decreases the required processing time but also increases the average recognition accuracy to 98.76% based on the Strathclyde dataset and to 98.21% based on the CapgMyo open source dataset. A relative comparison of experimental results between the proposed novel EMG based hand gesture recognition methodology and other similar approaches indicates the superior effectiveness of the new design.

Taking inspiration from the structure and behaviour of the human visual system and using the Transposed Convolution and Saliency Mapping methods of Convolutional Neural Networks (CNN), a spiking event-based image segmentation algorithm, SpikeSEG is proposed. The approach makes use of both spike-based imaging and spike-based processing, where the images are either standard images converted to spiking images or they are generated directly from a neuromorphic event driven sensor, and then processed using a spiking fully convolutional neural network. The spiking segmentation method uses the spike activations through time within the network to trace back any outputs from saliency maps, to the exact pixel location. This not only gives exact pixel locations for spiking segmentation, but with low latency and computational overhead. SpikeSEG is the first spiking event-based segmentation network and over three experiment test achieves promising results with 96% accuracy overall and a 74% mean intersection over union for the segmentation, all within an event by event-based framework.

Automatically detecting pathological voice disorders such as vocal cord paralysis or Reinke’s edema is an important medical classification problem. While deep learning techniques have achieved significant progress in the speech recognition field, there has been less research work in the area of pathological voice disorders detection. A novel system for pathological voice detection using Convolutional Neural Network (CNN) as the basic architecture is presented in this work. The novel system uses spectrograms of normal and pathological speech recordings as the input to the network. Initially Convolutional Deep Belief Network (CDBN) are used to pre-train the weights of CNN system. This acts as a generative model to explore the structure of the input data using statistical methods. Then a CNN is trained using supervised back-propagation learning algorithm to fine tune the weights. Results show that a small amount of data can be used to achieve good results in classification with this deep learning approach. A performance analysis of the novel method is provided using real data from the Saarbrucken Voice database.

In the last two decades, the use of synthetic aperture radar (SAR) for remote sensing purposes has significantly developed due to improvements in the quality and the availability of the images. Two powerful SAR techniques, namely, polarimetry and interferometry, have further increased the range of applications of the sensed data. Using polarimetry, geometrical properties and geophysical parameters, such as shape, roughness, texture, and moisture content, can be retrieved with considerable accuracy, while interferometric information may be used to extract vertical information with accuracy less than 1 cm. In this paper, the potential of using joint polarimetry and interferometry techniques in SAR data (PolInSAR) for the purpose of SAR image classification is investigated. To achieve this goal, we extend a covariance symmetry detection framework to the PolInSAR scenario. The proposed approach will be shown to be able to exploit the peculiar structures of the covariance matrices of PolInSAR images to discriminate structures within the image. Results using real-SAR data are presented to validate the effectiveness of the proposed approach.

The availability of multiple images of the same scene acquired with the same radar but with different polarizations, both in transmission and reception, has the potential to enhance the classification, detection, and/or recognition capabilities of a remote sensing system. A way to take advantage of the full-polarimetric data is to extract, for each pixel of the considered scene, the polarimetric covariance matrix, the coherence matrix, and the Muller matrix and to exploit them in order to achieve a specific objective. A framework for detecting covariance symmetries within polarimetric synthetic aperture radar (SAR) images is here proposed. The considered algorithm is based on the exploitation of special structures assumed by the polarimetric coherence matrix under symmetrical properties of the returns associated with the pixels under test. The performance analysis of the technique is evaluated on both simulated and real L-band SAR data, showing a good classification level of the different areas within the image.

This paper considers the problem of coherent (in the sense that both amplitudes and relative phases of the polarimetric returns are used to construct the decision statistic) multipolarization synthetic aperture radar change detection starting from the availability of image pairs exhibiting possible power mismatches/miscalibrations. The principle of invariance is used to characterize the class of scale-invariant decision rules which are insensitive to power mismatches and ensure the constant false alarm rate property. A maximal invariant statistic is derived together with the induced maximal invariant in the parameter space which significantly compresses the data/parameter domain. A generalized likelihood ratio test is synthesized both for the cases of two- and three-polarimetric channels. Interestingly, for the two-channel case, it is based on the comparison of the condition number of a data-dependent matrix with a suitable threshold. Some additional invariant decision rules are also proposed. The performance of the considered scale-invariant structures is compared to those from two noninvariant counterparts using both simulated and real radar data. The results highlight the robustness of the proposed method and the performance tradeoff involved.

Coherent multipolarization synthetic aperture radar (SAR) change detection exploiting data collected from N multiple polarimetric channels is addressed in this letter. The problem is formulated as a binary hypothesis testing problem, and a special block-diagonal structure for the polarimetric covariance matrix is forced to design a detector based on the generalized likelihood ratio test (GLRT) criterion. It is shown that the structured decision rule ensures the constant false alarm rate property with respect to the unknown disturbance covariance. Results on both simulated and real high-resolution SAR data show the effectiveness of the considered decision rule and its superiority against the traditional unstructured GLRT in some scenarios of practical interest.

This paper deals with coherent (in the sense that both amplitudes and relative phases of the polarimetric returns are used to construct the decision statistic) multipolarization synthetic aperture radar (SAR) change detection assuming the availability of reference and test images collected fromNmultiple polarimetric channels. At the design stage, the change detection problem is formulated as a binary hypothesis testing problem, and the principle of invariance is used to come up with decision rules sharing the constant false alarm rate property. The maximal invariant statistic and the maximal invariant in the parameter space are obtained. Hence, the optimum invariant test is devised proving that a uniformly most powerful invariant detector does not exist. Based on this, the class of suboptimum invariant receivers, which also includes the generalized likelihood ratio test, is considered. At the analysis stage, the performance of some tests, belonging to the aforementioned class, is assessed and compared with the optimum clairvoyant invariant detector. Finally, detection maps on real high-resolution SAR data are computed showing the effectiveness of the considered invariant decision structures.

The improved mean shift (IMS) algorithm can effectively track a colour target undergoing fast motion and complete occlusion. Instead template matching based on the sum of weighted absolute differences (SWAD) can track a target precisely, even under partial occlusion. To take advantage of both trackers, this paper presents a novel tracking algorithm that, on one side, further improves the IMS and extends it to a more general and flexible framework; on the other side, the proposed algorithm incorporates SWAD minimisation to increase the precision of the tracking results. Experimental results demonstrate that such an algorithm can robustly and precisely track targets with fast motion and in complete occlusion, more than conventional mean shift, IMS and SWAD-based trackers taken separately.

Classification of micro calcification clusters is very essential for early detection of breast cancer from mammograms. In this paper, an improved support vector machine (SVM) scheme is proposed, where optimized decision making is introduced for effective and more accurate data classification. Experimental results on the well-known DDSM database have shown that the proposed method can significantly increase the performance in terms of F1 and Az measurements for the successful classification of clustered micro calcifications.

Synthetic aperture radar (SAR) systems are used to form high-resolution images from radar backscatter signals. The fractional Fourier transform (FrFT), which is a generalised form of the well-known Fourier transform, has opened up the possibility of a new range of potentially promising and useful applications that involve the use and detection of chirp signals that include pattern recognition and SAR imaging. In this study a time variant problem associated with the use of the FrFT for SAR processing is addressed and a new algorithm is presented that resolves this problem. Two new FrFT-based SAR processing algorithms are presented namely the FrRDA and the eFrCSA that are shown to improve the well-established range-Doppler and chirp-scaling algorithms for SAR processing. The performance of the algorithms are assessed using simulated and real Radarsat-1 data sets. The results confirm that the FrFT-based SAR processing methods provide enhanced resolution yielding both lower side lobes effects and improved target detection.

The effectiveness of frequency domain processing algorithms in bistatic synthetic aperture radar (SAR) focusing depends critically on the accuracy of the bistatic slant range function approximation. This letter presents a new Chebyshev slant range function approximation that is shown to increase the accuracy of the analytical approximation of the bistatic point target spectrum. The performance of the new method is compared to the conventional Taylor series approximation approach in the generation of the point target spectrum. The new approach is shown to provide a more accurate approximation of the slant range function with negligible increase in processing requirements compared to the traditional Taylor series approximation. The accuracy improvement is shown to yield a more accurate spectrum that can be exploited in bistatic SAR focusing algorithms.

An empirical mode decomposition-based filtering (EMDF) approach is presented as a postprocessing stage for speech enhancement. This method is particularly effective in low-frequency noise environments. Unlike previous EMD-based denoising methods, this approach does not make the assumption that the contaminating noise signal is fractional Gaussian noise. An adaptive method is developed to select the IMF index for separating the noise components from the speech based on the second-order IMF statistics. The low-frequency noise components are then separated by a partial reconstruction from the IMFs. It is shown that the proposed EMDF technique is able to suppress residual noise from speech signals that were enhanced by the conventional optimally modified log-spectral amplitude approach which uses a minimum statistics-based noise estimate. A comparative performance study is included that demonstrates the effectiveness of the EMDF system in various noise environments, such as car interior noise, military vehicle noise, and babble noise. In particular, improvements up to 10 dB are obtained in car noise environments. Listening tests were performed that confirm the results.

Bistatic synthetic aperture radar (BSAR) provides strategical, technical and economical advantages in radar imaging. Motions and micro-motions of objects in an illuminated scene introduces Doppler and micro-Doppler effects in the received radar echoes. Combining the advantages introduced by the bistatic configuration and the usefulness of the micro-Doppler signature characterization will provide a powerful tool for military and civil remote sensing applications such as target recognition and classification. In this paper, a vibrating micro-Doppler signature for a BSAR system with fixed receiver is analyzed and compared to the signature obtained in a monostatic SAR system. The micro-Doppler effect is derived for a vibrating target in the bistatic SAR. The corresponding bistatic factor is shown to be a function of the bistatic acquisition geometry. Also, the effect of the target vibration on the focused image is shown to be influenced by the acquisition geometry. The derived model is useful for micro-Doppler classification. Simulations for 94 GHz and 10 GHz are given and the results confirm the derived model.

This paper presents a new feature extraction technique called Pulse Active Mean (PAM) implemented on Electrocardiograms (ECG) for biometric authentication. A doubly secure ECG authentication framework is proposed which makes use of the important attributes of the PAM algorithm as a personal identification number (PIN). The PIN is used to extract different locations of ECG characteristics generating unique feature vectors. The presence of the correct PIN and ECG signals make the proposed authentication framework doubly secure. The performance of PAM is evaluated by comparing its receiver operating characteristic (ROC) curve with traditional temporal and amplitude feature extraction techniques on 100 Physikalisch-Technische Bundesanstalt (PTB) subjects. The evaluation of the biometric performance when different values of PIN are presented is also investigated. It is shown in this paper that different PIN values generate different feature vector sets while still providing consistent authentication performance.

The effectiveness of the conventional Mean Shift tracking algorithm diminishes for fast moving targets and complete occlusion. In this paper an improved Mean Shift algorithm comprising a fast failure recovery strategy that aims to deal with randomly moving targets and complete occlusion as encountered in crowded scenes is presented. Experimental results show that after complete occlusion or target loss, the new algorithm can effectively recover and continue to successfully track targets in complex scenarios.

This letter presents a new signal processing subsystem for conventional monopulse tracking radars that offers an improved solution to the problem of dealing with manmade high power interference (jamming). It is based on the hybrid use of empirical mode decomposition (EMD) and fractional Fourier transform (FrFT). EMD-FrFT filtering is carried out for complex noisy radar chirp signals to decrease the signal's noisy components. An improvement in the signal-to-noise ratio (SNR) of up to 18 dB for different target SNRs is achieved using the proposed EMD-FrFT algorithm.

Biometric authentication is a one-to-one verification process. A successful biometric system must be capable of avoiding a fraudulent claim while at the same time ensuring the privacy of individuals. This paper presents a novel framework for using the electrocardiogram (ECG) as a biometric system for human authentication. A new feature extraction technique known as pulse active ratio (PAR) is derived and used to generate novel ECG feature vectors. The proposed method is validated by experiments on 112 subjects, performing 9800 ECG comparisons providing a 10% improvement when compared to conventional temporal and amplitude feature extraction methods.

Conventional monopulse radar processors are used to track a target that appears in the look direction beam width. The distortion produced when additional targets appear in the look direction beam width can cause severe erroneous outcomes from the monopulse processor. This leads to errors in the target tracking angles that may cause the target tracker to fail. A new signal processing algorithm is presented in this paper that is based on the use of optimal Fractional Fourier Transform (FrFT) filtering to solve this problem. The relative performance of the new filtering method over traditional based methods is assessed using standard deviation angle estimation error (STDAE) for a range of simulated environments. The proposed system configurations with the optimum FrFT filters succeeds in effectively cancelling additional target signals appearing in the look direction beam width.

In this paper we present a new target tracking enhancement system that uses a Kalman filter in the presence of interference. If the radar (seeker) is affected by different types of interference, this will affect the missile trajectory towards the target and may cause inaccurate tracking. In the new system a six-state Kalman filter is utilized to perform the tracking task and to carry out smoothing to the corrupted trajectory. This also provides good information about the target velocity in three dimensions which is very important information about the target. A three dimensional scenario between target (with high manoeuvre) and missile is used to illustrate the performance of the system in the case when (i) no interference is present and (ii) interference is present. The performance of the filtered trajectory using the Kalman tracker will be assessed for different guidance methods: including (i) proportional navigation (ii) pure pursuit and (iii) constant bearing. The Kalman improvement for the tacking for the three guidance method will be analysed.

Sprite generation is extensively used for the video compression. Real time efficient video analytics (VA) for surveillance system need to make decision with in camera processing unit or at camera edge processing unit. The Video Analytics is required to work on good resolution footage to register tiny but important information that may otherwise be lost during subsequent compression process, especially in temporally exploited streams. This paper presents a novel approach to generate a full view panoramic mosaic for a surveillance sequence to help detect moving objects and extract MPEG-7 motion descriptors. Unlike the current panoramic transformation methods, in which pure horizontal panning motion is requiredis required to construct the mosaic, our method can deal with uncontrolled panning motion with slight camera tilt motion. This method provides a generic solution for real world surveillance camera issues such as undesired tilt motion because of the nonlinearity of camera moving mechanism, strong wind etc. The sprite generated helps to detect moving objects, constructing motion trajectories and also provides a clean background reference even for a panning camera. The algorithm has been evaluated with several test sequences.

This paper proposed a novel approach based on Local Linear Embedding (LLE) for modeling and understanding the temporal behaviour from facial paralysis image sequences. LLE is sensitive to scaling, illumination and face pose. A Supervised LLE (SLLE) based on directed Hausdorff distance is proposed for aligning two image sets from the two sides of the face using one generalized embedding space. The embedded low dimensional manifolds represent the asymmetry of the facial motion regardless of the extrinsic facial asymmetry caused by the natural bilateral asymmetry, illumination and shadows. Experimental results demonstrate that our approach is more reliable than frame difference based methods and optical flow based methods without a significant increase in computational complexity.

This paper presents a robust, objective, automated and quantitative assessment system for facial paralysis using artificial intelligence analysis of biomedical video data. Facial feature localization and prescribed facial movements detection are discussed. Optical flow is used to obtain the motion features in the relevant facial regions. Radial basis function (RBF) neural network is applied to provide quantitative evaluation of facial paralysis based on the House-Brackmann scale. The results from 197 videos of 87 subjects are encouraging with a mean squared error (MSE) of 0.013 (training) and 0.0169 (testing).

The fractional Fourier transform (FrFT), which is a generalized form of the well-known Fourier transform, has opened up the possibility of a new range of potentially promising and useful applications including radar involving the use and detection of chirp signals, pattern recognition and Synthetic Aperture Radar (SAR) image processing. The Chirp Scaling Algorithm (CSA) is one of the most important and well-known radar imaging algorithms. It is attractive because of its excellent focusing ability and implementation simplicity. Benefiting from the inherent structure of the FrFT for non- stationary digital signal processing and analysis, especially for chirped-type signals, a new version of the CSA based on the Fractional Fourier Transform (FrFT) is developed. The introduced Fractional Chirp Scaling Algorithm (FrCSA) applied the Fast Fourier Transform (FFT) instead of the fractional Fourier transform (FrFT) in the azimuth direction for the analytical development tractability purposes only as it numerically tractable in both dimensions. To demonstrate the resolution and focusing enhancement in the azimuth dimension using the FrCSA and also to be able to perform azimuth fractional filtering, noise removal and flight path nonlinearity compensation, a closed form expression for the azimuth fractional transformation is required. In this talk we present the mathematical derivation for a closed- form expression of the azimuth-fractional Fourier transform of the new FrCSA with application to high resolution imaging. Results to real SAR data images will show significantly enhanced features using the FrFT-based azimuth expression instead of the classical FFT-based one within the fractional chirp scaling algorithm or any other chirped-type SAR imaging algorithm.

The fractional Fourier transform (FrFT), which is a generalized form of the well-known Fourier transform, has only recently started to appear in the field of signal processing. This has opened up the possibility of a new range of potentially promising and useful applications. In this paper we apply the new FrFT-based chirp scaling algorithm (CSA) to a high resolution-high focused synthetic aperture radar (SAR) imaging and compare its performance with the classical CSA based on the fast Fourier transform (FFT). Simulation results show that the FrFT-based CSA can offer significantly enhanced features compared to the classical FFT-based approach

In this paper a new adaptive subband thresholding algorithm is introduced for finding the optimize threshold values of the subbands. The search technique is exhaustive, where the distortion level and bit allocation are used as the limiting factors. The algorithm is applied in multistage lattice VQ (MLVQ) to enhance the PSNR performance. The MLVQ image coding technique reduces the quantization error of the quantized vectors by "blowing out" the residual quantization errors with a LVQ scale factor. Experimental results for images using the adaptive subband thresholding technique and the MLVQ are shown to be superior to those of JPEG2000 for a set of test images

The fractional chirp scaling algorithm (FrCSA) is based on the use of the fractional Fourier transform (FrFT) within the chirp scaling algorithm (CSA). In this paper, a closed-form expression for the azimuth FrFT of the FrCSA is mathematically derived and analyzed from the high-resolution synthetic aperture radar imaging point of view. The azimuth-FrFT expression of the FrCSA is compared to that of the classical fast Fourier transform (FFT)-based CSA. As the FFT is a special case of the generalized FrFT, the derived expression is found to be in total agreement with that of the FFT-based CSA when the transformation order is equal to unity; that is the angle of rotation is equal to pi/2

Fresnelets are wavelet-like base functions specially tailored for digital holography applications. We introduce their use in phase-shifting interferometry (PSI) digital holography for the compression of such holographic data. Two compression methods are investigated. One uses uniform quantization of the Fresnelet coefficients followed by lossless coding, and the other uses set portioning in hierarchical trees (SPIHT) coding. Quantization and lossless coding of the original data is used to compare the performance of the proposed algorithms. The comparison reveals that the Fresnelet transform of phase-shifting holograms in combination with SPIHT or uniform quantization can be used very effectively for the compression of holographic data. The performance of the new compression schemes is demonstrated on real PSI digital holographic data.

Facial paralysis is a debilitating condition in which sufferers experience unilateral paralysis of the left or right facial nerve. An evidence based assessment of a patient's condition is almost impossible because all current grading scales are subjective. A quantitative, practical, reliable system would be an invaluable tool in this field of neurootology. Demonstrated here is a system which intelligently quantifies the facial damage in 43 testing videos from 14 subjects. Using an artificial neural network the average mean squared error for the system is 1.6%.

The fractional Fourier transform (FrFT), which is a generalized form of the well-known Fourier transform, has only recently started to appear in the field of signal processing. This has opened up the possibility of a new range of potentially promising and useful applications. In this letter, we develop a new FrFT-based chirp scaling algorithm (CSA) and compare its performance with the classical CSA based on the fast Fourier transform (FFT). Simulation results show that the FrFT-based CSA can offer significantly enhanced features compared to the classical FFT-based approach.

Sea clutter in marine surveillance radar makes the task of detecting small targets a very challenging problem. In this paper, a set of three signal processing techniques designed to suppress unwanted sea clutter radar echo and achieve target detection with no prior knowledge of the ocean and environmental conditions is presented. These include signal averaging, time-frequency representation, and morphological filtering. Datasets from real marine radar operating in staring mode are used to illustrate the performance of the new approaches.

This paper proposes the use of a polynomial interpolator structure (based on Horner's scheme) which is efficiently realizable in hardware, for high-quality geometric transformation of two- and three-dimensional images. Polynomial-based interpolators such as cubic B-splines and optimal interpolators of shortest support are shown to be exactly implementable in the Horner structure framework. This structure suggests a hardware/software partition which can lead to efficient implementations for multidimensional interpolation.

This paper is concerned with the classification and de-noising problem for non-linear signals. It is known that using kernel functions, a non-linear signal can be transformed into a linear signal in a higher dimensional space. In that feature space, a linear algorithm can be applied to a non-linear problem. It is proposed that using the principal components extracted from the feature space, the signal can be classified correctly in its input space. Additionally, it is shown how this classification process' can be used to de-noise DQPSK communication signals.

A general class of Computationally Efficient locally Recurrent Networks (CERN) is described for real-time adaptive signal processing. The structure of the CERN is based on linear-in-the-parameters single-hiddenlayered feedforward neural networks such as the Radial Basis Function (RBF) network, the Volterra Neural Network (VNN) and the recently developed Functionally Expanded Neural Network (FENN), adapted to employ local output feedback. The corresponding learning algorithms are described and key structural and computational complexity comparisons are made between the CERN and conventional Recurrent Neural Networks. A speech signal is used, which shows that a Recurrent FENN based adaptive CERN predictor can significantly outperform the corresponding feedforward FENN and conventionally employed linear adaptive filtering models.

One of the main problems associated with object based video coding systems, is how to efficiency extract objects. MPEG-4 is an object based system using video object planes (VOPs), however MPEG-4 does not define how VOPs are extracted from video sequences. This article presents a human visual system based technique for extracting objects, that combines motion segmentation, image segmentation, and a contrasted image to extract objects, that can be applied to VOP extraction.

In MPEG-4 that uses video object planes (VOPs), systems to efficiently extract objects are required. Based on research into the human visual system (HVS), this paper presents an object extraction system that uses motion segmentation and image segmentation in parallel, and then combines them to efficiently extract objects using fuzzy reasoning. The most computationally demanding part of this system, the motion segmentation (MS), is implemented on the TMS320C80 DSP MVP. The motion estimation part of the MS algorithm and its fast implementation on C80 architecture are also described.

A new approach for the decision feedback equalizer (DFE) based on the functional-link neural network is described. The structure is applied to the problem of adaptive equalization in the presence of intersymbol interference (ISI), additive white Gaussian noise, and co-channel interference (CCI). It is shown through simulation results for a severe amplitude distorted co-channel system that the decision feedback functional-link equalizer (DFFLE) provides significantly superior bit-error rate (BER) performance characteristics compared to the conventional DFE, the linear transversal equalizer (LTE), the nonlinear radial basis function (RBF) neural-network-based structures and the feed-forward functional-link equalizer (FFLE)-based structures. The DFFLE is also shown to have a significantly simpler computational requirement relative to the RBF and the FFLE.

A novel adaptive vector quantization (VQ) based subband coding scheme for very low bit rate coding of video sequences is presented. Overlapped block motion estimation/compensation is employed to exploit interframe redundancy. A two-dimensional (2-D) wavelet transform (WT) is applied to the resulting displaced frame difference (DFD) signal. The WT coefficients are encoded using an adaptive subband vector quantization (ASBVQ) scheme in combination with a dynamic bit allocation strategy based on marginal analysis. Fixed rate coding is provided. Comparative experimental results of the ASBVQ codec with the H.261 and the H.263 video coding standards are given.

This paper brings together two strands of current interest in signal processing. While second order techniques and minimum variance criteria are well understood, there is a growing requirement to study the performance criteria that involve higher order statistics in order to evaluate deviation from Gaussianity, linearity and stationarity of observed data. There is a complimentary requirement for the generation of random test sequences which have prescribed (or minimal) higher order statistics, to facilitate the analysis of systems in order to determine linearity/non-linearity, time invariance vs time varying parameters. This paper proposes a new method for minimising the third order cumulant spread of random sequences with symmetric pdf, and provides a closed form solution for the weightings required to achieve this. Numerous computed results are included to verify performance.>

This paper presents a study of a new class of adaptive channel equaliser, known as the functional-link (FL) equaliser which utilises functional expansion of the equaliser's input data. By carefully selecting the appropriate functions of the input, significant performance improvement can be obtained. These studies will lead to significant reduction in the exponentially increased expansion of the polynomial-perceptron equaliser. Numerical simulation results are presented to highlight the better bit error rate performance of the FL based equaliser compared to other non-linear, neural network based equalisers.>

Novel wavelet transform based schemes for coding still images and image sequences are presented. The still image codec uses a new efficient adaptive bit-plane run-length coding of the wavelet transform coefficients of images. The main merit of this coding scheme is its simplicity requiring no training or storage of codebooks and it outperforms the JPEG at low bit-rate. For image sequence coding, a very low bit-rate sub-band motion estimation/compensation video codec designed for colour video conferencing applications is presented. A full system with buffer feedback control is designed. Simulation results of transmitting colour image sequences at 9.6 to 19.2 kbit/s are given.>

The stability of the noncanonical least mean square (NCLMS) algorithm is investigated. The NCLMS effectively uses a different step size for each tap coefficient position during adaptation. The classical LMS step size bound cannot be directly applied to the NCLMS. The weight error vector is modeled as a first-order difference equation and a stability bound for the NCLMS is derived. Simulation results are presented to back up the analysis.>

The authors present analysis and simulations of an LMS (least mean square) based adaptive filtering algorithm called the NCLMS (non-canonical LMS). Rather than using the standard FIR (finite impulse response) filter as for the LMS algorithm, a modified structure called the NCFIR (non-canonical FIR) is used. The NCFIR allows a faster VLSI implementation than the conventional FIR. A comparison of the performances of the NCLMS and conventional LMS algorithm is presented for an inverse system modeling application. Simulation results are given which show a reduced EMSE (excess mean square error) level and an improved performance in an impulsive noise environment for the NCLMS over the LMS algorithm.>

A computationally efficient algorithm for direction finding is described. The sample covariance matrix is averaged into d column vectors, where d is the known number of sources. These column vectors are used as a basis for the signal subspace. They can be used directly for the two subspace estimates in the ESPRIT algorithm or they can be used to construct a projection matrix for use with the MUSIC algorithm. For low signal-noise-ratio conditions, a subtraction process is used to remove the diagonal noise terms from the covariance matrix and as a result a shorter search vector is used in the MUSIC algorithm.>

A parallel implementation of the two-dimensional, adaptive blocksize discrete cosine transform (DCT) is based on the principle of the quadtree structure. An efficient coding technique for the quadtree used is described. A network of Inmos T800 transputers is chosen as the target system. Two types of parallel network topologies are analyzed, and the results for execution time are presented. The results show that the tree structure is faster than the pipelining ring structure for this particular application.>

The maximum eigenvalue varied rate criteria (MEVRC) method is presented as a novel approach for determining the number of harmonics in a given signal. In contrast to the AIC and MDL criteria, MEVRC does not use a bias term. It is based on the eigenvalue varied rate of the smallest eigenvalues to estimate the number of signals. Simulation results indicate that the performance of MEVRC is better than AIC in normal signal-to-noise ratio (SNR) and sample situations, and better than the MDL in adverse conditions.>

A fast implementation of the ESPRIT algorithm, incorporating parallel processing, is described. ESPRIT is a novel, subspace fitting, parameter estimation algorithm which is used for obtaining high-resolution, unbiased estimates of the frequencies and powers of complex sinusoids in noise. The algorithm involves complex decompositions which require extensive computation. In order to use the algorithm in close to real-time situations, it is altered to facilitate faster computation and the use of a parallel architecture. The altered algorithm is decomposed onto an array of Inmos T800 transputers and its performance compared to that of the standard TLS-ESPRIT through simulations.>

A method of shape recognition involving fast correlation techniques based on number theoretic transforms (NTTs) is outlined. NTTs facilitate the computation of convolutions with no roundoff error and are ideally suited for implementation on current bit-serial word-parallel machines such as the DAP-500. The transforms can be carried out by a combination of shifts and adds with all arithmetic performed modulo some integer number. They produce very fast implementations at the price of limiting the input to integer values over a finite field. In the present application the pixel values describing an object boundary are in precisely this form so that the advantages of the NTT can be obtained at no cost in quantization or clipping noise. Results of correlating shapes which have undergone various transformations are presented.>

The performance of a parallel algorithm relies heavily on the routing demands of the algorithm and how efficiently the parallel system copes with the demands. Highly parallel, bit-serial machines became most powerful when the application involves low precision or integer data. Lowering the precision results in an increase in algorithmic noise. Algorithms are presented that implement the noiseless data permutations required in the prime factor and Winograd's discrete Fourier transform algorithms on a highly parallel processor array that has parallel access only to bit planes. The costs for both the conventional 'twiddles' and for the new algorithms are included using the ICL, 200 ns, distributed array processor (DAP).>

After first outlining the principles of Synthetic Aperture Radar (SAR) and reviewing the main features of the ICL Distributed Array Processor (DAP), efficient DAP algorithms are described for fast azithmuth correlation of range compressed SAR signals. Range migration correction and speckle reduction by multilook processing are considered. Typical timing results are presented for correlation of simulated SAR signals from SEASAT and ERS-1 systems using 24-bit floating point arithmetic.