No description available.

An acoustic array was deployed in the nearsurface layer in Saanich Inlet, B.C. to image breaking waves using only the naturally occurring acoustical radiation from the breaking region over the band [160 Hz, 2000 Hz]. The 1.5-element array was configured as a horizontal cross with an 8 m aperture, bottom-moored, and positioned nominally 3 m beneath the surface. Due to sensor sparseness, the array PSF at any particular frequency was badly contaminated by grating lobes. A novel broadband scheme was devised to combine information at multiple independent frequencies to yield unambiguous images with resolution of about 0.2 m at the sea surface. The broadband scheme assumed space-time separability in the source mutual spectral density. This is only considered valid for breaking waves above about 400 Hz. Nonstationarity and time-bandwidth constraints yielded at most six independent frequency bands within the system passband. A parametric image analysis showed that the images align closely with the wind and can be observed moving downwind with a speed about two-thirds the phase speed of the dominant component of the wind waves. Absolute power levels were found to be consistent with previously published results. The absolute power levels were parameterized by where [special characters omitted] and λ (f) is well-described by a simple first order relation [special characters omitted], where [special characters omitted] varied depending on the size of the wave but b1 appeared to be a more universal constant estimated at -4.55 ± 0.47. The source mechanism for frequencies below about, 400 Hz was modeled two ways: (1) as a point source (which would follow if an acoustically compact “collective oscillation" region had formed), and (2) as due to off-peak spectral contributions from bubbles resonant at 400 Hz. Neither model achieved a satisfactory fit to the observed data. This seems to imply that the mechanism below about 400 Hz was acoustically extended and radiating as energetically as any resonant bubbles. / Graduate

This study proposes a generic model for optical character recognition (OCR) on remote objects, using digitised video and digital camera input. Currently, most applications using this technique are ad hoc. Lack of formalism creates the risk of fragmented adhocracy. The purpose of this model is to introduce some degree of formalism based on classical character recognition, thereby providing a theoretical basis, and to explore the potential of character recognition on digital images for automated reading. The model also propogates the use of off the shelf commercial software and hardware currently available. The model “CRODI” which is an abbreviation for Character Recognition On Digital Images, consists of several objects or building blocks necessary for any remote character recognition application. In order to provide an optimal solution for any application it is necessary to select the most appropriate and effective building blocks. The combination and sequence in which these building blocks are selected is called a visual routine. This study does not only focus on choosing the correct building blocks for an application but also describes how certain visual routines and methodologies can be used to cater for applications with a certain profile. Furthermore, a formal notation in the form of an Expert Support System is introduced to describe optimised visual routines and the selection of building blocks in an efficient way. An example is given in chapter six of the efficiency and effectiveness of the model on a classical but dynamic application, where character recognition is performed on stationary and fast moving vehicles’ number plates. The biggest advantage of CRODI is to propose a generic framework or model from which an intuitive selection of objects, visual routines or methodologies can occur to expedite and optimise remote character recognition on almost any stationary or moving object. / Ehlers, E.M., Prof.

This thesis summarises research on the improved design, integration and expansion of past cortex-like computer vision models, following biologically-inspired methodologies. By adopting early theories and algorithms as a building block, particular interest has been shown for algorithmic parameterisation, feature extraction, invariance properties and classification. Overall, the major original contributions of this thesis have been: 1. The incorporation of a salient feature-based method for semantic feature extraction and refinement in object recognition. 2. The design and integration of colour features coupled with the existing morphological-based features for efficient and improved biologically-inspired object recognition. 3. The introduction of the illumination invariance property with colour constancy methods under a biologically-inspired framework. 4. The development and investigation of rotation invariance methods to improve robustness and compensate for the lack of such a mechanism in the original models. 5. Adaptive Gabor filter design that captures texture information, enhancing the morphological description of objects in a visual scene and improving the overall classification performance. 6. Instigation of pioneering research on Spiking Neural Network classification for biologically-inspired vision. Most of the above contributions have also been presented in two journal publications and five conference papers. The system has been fully developed and tested in computers using MATLAB under a variety of image datasets either created for the purposes of this work or obtained from the public domain. / © Cranfield University

University of Macau / Faculty of Science and Technology / Department of Computer and Information Science

Please read the abstract in the section 00front of this document / Thesis (PhD (Electronic Engineering))--University of Pretoria, 2007. / Electrical, Electronic and Computer Engineering / unrestricted

Humans spend a large amount of their time interacting with computers of one type or another. However, computers are emotionally blind and indifferent to the affective states of their users. Human-computer interaction which does not consider emotions, ignores a whole channel of available information. Faces contain a large portion of our emotionally expressive behaviour. We use facial expressions to display our emotional states and to manage our interactions. Furthermore, we express and read emotions in faces effortlessly. However, automatic understanding of facial expressions is a very difficult task computationally, especially in the presence of highly variable pose, expression and illumination. My work furthers the field of automatic facial expression tracking by tackling these issues, bringing emotionally aware computing closer to reality. Firstly, I present an in-depth analysis of the Constrained Local Model (CLM) for facial expression and head pose tracking. I propose a number of extensions that make location of facial features more accurate. Secondly, I introduce a 3D Constrained Local Model (CLM-Z) which takes full advantage of depth information available from various range scanners. CLM-Z is robust to changes in illumination and shows better facial tracking performance. Thirdly, I present the Constrained Local Neural Field (CLNF), a novel instance of CLM that deals with the issues of facial tracking in complex scenes. It achieves this through the use of a novel landmark detector and a novel CLM fitting algorithm. CLNF outperforms state-of-the-art models for facial tracking in presence of difficult illumination and varying pose. Lastly, I demonstrate how tracked facial expressions can be used for emotion inference from videos. I also show how the tools developed for facial tracking can be applied to emotion inference in music.

This thesis presents a theory of multi-scale, curvature and torsion based shape representation for planar and space curves. The theory presented has been developed to satisfy various criteria considered useful for evaluating shape representation methods in computer vision. The criteria are: invariance, uniqueness, stability, efficiency, ease of implementation and computation of shape properties. The regular representation for planar curves is referred to as the curvature scale space image and the regular representation for space curves is referred to as the torsion scale space image. Two variants of the regular representations, referred to as the renormalized and resampled curvature and torsion scale space images, have also been proposed. A number of experiments have been carried out on the representations which show that they are very stable under severe noise conditions and very useful for tasks which call for recognition of a noisy curve of arbitrary shape at an arbitrary scale or orientation. Planar or space curves are described at varying levels of detail by convolving their parametric representations with Gaussian functions of varying standard deviations. The curvature or torsion of each such curve is then computed using mathematical equations which express curvature and torsion in terms of the convolutions of derivatives of Gaussian functions and parametric representations of the input curves. Curvature or torsion zero-crossing points of those curves are then located and combined to form one of the representations mentioned above. The process of describing a curve at increasing levels of abstraction is referred to as the evolution or arc length evolution of that curve. This thesis contains a number of theorems about evolution and arc length evolution of planar and space curves along with their proofs. Some of these theorems demonstrate that evolution and arc length evolution do not change the physical interpretation of curves as object boundaries and others are in fact statements on the global properties of planar and space curves during evolution and arc length evolution and their representations. Other theoretical results shed light on the local behavior of planar and space curves just before and just after the formation of a cusp point during evolution and arc length evolution. Together these results provide a sound theoretical foundation for the representation methods proposed in this thesis. / Science, Faculty of / Computer Science, Department of / Graduate

Markov Random Fields (MRFs) are used in computer vision as an effective method for reconstructing a function starting from a set of noisy, or sparse data, or in the integration of early vision processes to label physical discontinuities. The MRF formalism is attractive because it enables the assumptions used to be explicitly stated in the energy function. The drawbacks of such models have been the computational complexity of the implementation, and the difficulty in estimating the parameters of the model. In this thesis, the deterministic approximation to the MRF models derived by Girosi and Geiger[10] is investigated, and following that approach, a MIMD based algorithm is developed and implemented on a network of T800 transputers under the Trollius operating system. A serial version of the algorithm has also been implemented on a SUN 4 under Unix. The network of transputers is configured as a 2-dimensional mesh of processors (currently 16 configured as a 4 x 4 mesh), and the input partitioning method is used to distribute the original image across the network. The implementation of the algorithm is described, and the suitability of the transputer for image processing tasks is discussed. The algorithm was applied to a number of images for edge detection, and produced good results in a small number of iterations. / Science, Faculty of / Computer Science, Department of / Graduate

M. Ing. / The purpose of this study is to investigate the use of computer vision techniques to measure the electrode slip. The study investigates a potential location for camera placement in the furnace housing, as well as the use of computer vision algorithms that could be used to solve the problem. A slip measurement algorithm is then designed, implemented and tested. The implemented slip measurement algorithm is based on the manual slip measurement technique, by measuring relative electrode and slip arm displacement between the electrode and the slip arm. The algorithm uses SURF invariant features to extract the electrode features and slip arm features in one frame, and match these features to the next frame SURF. Scene calibration is then used to relate the pixel slip measurement to a metric distance measurement. The experimental results proved that there is scope for applying computer vision techniques to address the slip measurement problem, using a single HD camera. However, there is room for improvement and the recommendations and future work are also discussed.