This research focuses on the development of a novel image super-resolution algorithm for coordinate measurement in manufacturing. The main features of the algorithm are that it is fast, flexible and fully automatic. A fast algorithm is required because image-super resolution is a procedure that handles a large amount of data. Having a slow or highly complex algorithm may result in computational infeasibility. A flexible algorithm means the algorithm can be customised to handle specific problems, i.e. the algorithm can be augmented with multiple constraints and still obtain an optimal solution. This is desirable as most image super-resolution problems are specific and having the capacity to augment multiple constraints reduces the search space, thus leading to faster convergence. An automatic algorithm is viewed as ideal as it has minimum human intervention and will generate super-resolution images automatically when measured frames are input. / This study considers three issues related the developing the algorithms: the model of image super-resolution; the formulation of a flexible algorithm that is capable of augmenting multiple constraints into the model and produces optimal super-resolution images; and the optimisation technique to solve the problem formulated to ensure that the computational complexity is low. / Thesis (PhDEngineering)--University of South Australia, 2005.

No description available.

No description available.

No description available.

Bibliography: leaves 100-110. / xiv, 110 leaves : ill. ; 30 cm. / Title page, contents and abstract only. The complete thesis in print form is available from the University Library. / This thesis addresses the structure from motion problem in computer vision. / Thesis (Ph.D.)--University of Adelaide, Dept. of Computer Science, 2000

Typescript (photocopy) / vii, 190 leaves ; 30 cm. / Title page, contents and abstract only. The complete thesis in print form is available from the University Library. / Thesis (Ph.D.) Dept. of Electrical and Electronic Engineering, University of Adelaide, 1979

Adaptive signal processing has evolved in the last thirty years to the point where its use in sensors such as radar and sonar and in communications is indispensable. High frequency (HF) skywave radars benefit in particular from spatial and spatio-temporal adaptive filters, detectors and estimators due to their operation in an environment which is crowded with natural and man-made interferences, as well as significant temporal and spatial distortions due to ionospheric propagation. While adaptive processing is important for other types of sensors, including airborne radars, HF radar systems are particularly well-suited to its application, given the modern digital receiver-per-element arrays and radar facilities able to host large computational resources. This allows use of algorithms viewed as merely theoretical benchmarks for other systems. / However, despite the tremendous advances in radar adaptive signal processing theory since its foundation in the 1960s, a number of important issues have still not been addressed fully. In particular, the problem of limitations in available training data for adaptive estimation has, if anything, become more acute in recent years. In the case of HF radar, the hundreds of degrees of freedom presented by the typical HF array prevent the application of conventional techniques, not because of computational cost, but due to insufficient training sample support. Furthermore, new architectures for next generation systems including two-dimensional transmit and receive antenna arrays with MIMO technology to support non-causal adaptivity on transmit will further increase the demand for training data, making an already significant problem even more important in the future. / The following broad problems are found to be the most important at this stage: Without a prior knowledge of particular radar scenarios, how can the suitability of its adaptively reconstructed model for an associated radar inference be verified; what are the ultimate capabilities of adaptive techniques in the pre-asymptotic domain, beyond which the adaptive detection/ estimation problem cannot provide a consistent solution, and how can that limit be assessed in the absence of defined exact finite-sample statistical properties or by resorting to standard large-sample asymptotics; given a limited training data volume, what is this mix of credible a priori assumptions (parametric models) regarding this radar scenario, on one hand, and its adaptive estimation on the other? / Clearly each of these major questions is too complex to be comprehensively addressed in a single study. But this thesis (and the associated publications), by providing further understandings in each of these areas, introduces important results to the field of adaptive processing in the presence of low training sample support. / Thesis (PhDTelecommunications)--University of South Australia, 2009

No description available.

No description available.

A range of signal processing techniques have been adopted and developed as a methodology which can be used in developing an intelligent surface electromyography (SEMG) signal classifier. An intelligent SEMG signal classifier would be used for recognising and treatment of musculoskeletal pain and some neurological disorders by physiotherapists and occupational therapists. SEMG signals displays the electrical activity from a skeletal muscle which is detected by placing surface electrodes placed on the skin over the muscle. The key factors of this research were the investigation into digital signal processing using various analysis schemes and the use of the Artificial Neural Network (ANN) for signal classification of normal muscle activity. The analysis schemes explored for the feature extraction of the signals were the Fast Fourier Transform (FFT), Short Time Fourier Transform (STFT), Continuous Wavelet Transform (CWT), Discrete Wavelet Transform (DWT) and Discrete Wavelet Packet Transform (DWPT).Traditional analysis methods such as FFT could not be used alone, because muscle diagnosis requires time-based information. CWT, which was selected as the most suitable for this research, includes time-based information as well as scales, and can be converted into frequencies, making muscle diagnosis easier. CWT produces a scalogram plot along with its corresponding frequency-time based spectrum plot. Using both of these plots, overviewed extracted features of the dominant frequencies and the related scales can be selected for inputs to train and validate an ANN. The purpose of this research is to classify (SEMG) signals for normal muscle activity using different extracted features in an ANN. The extracted features of the SEMG signals used in this research using CWT were the mean and median frequencies of the average power spectrum and the RMS values at scales 8, 16, 32, 64 and 128. SEMG signals were obtained for a 10 second period, sampled at 2048 Hz and digitally filtered using a Butterworth band pass filter (5 to 500 Hz, 4th order). They were collected from normal vastus lateralis and vastus medialis muscles of both legs from 45 male subjects at 25%, 50%, and 75% of their Maximum Voluntary Isometric Contraction (MVIC) force of the quadriceps. The ANN is a computer program which acts like brain neurons, recognises, learns data and produces a model of that data. The model of that data becomes the target output of an ANN. Using the first 35 male subjects' data sets of extracted features, the ANN was trained and then validated with the last 10 male subjects' data sets of the untrained extracted features. The results showed how accurate the untrained data were classified as normal muscle activity. This methodology of using CWT for extracting features for analysing and classifying by an ANN for SEMG signals has shown to be sound and successful for the basis implementation in developing an intelligent SEMG signal classifier.