W. Eric L. Grimson ; with contributions from Tomá³ Lozano-Pé²¥z, Daniel P. Huttenlocher. / Includes bibliographical references (p. 350-504) and indexes.

Thesis (MScEng)--Stellenbosch University, 2003. / ENGLISH ABSTRACT: We investigate object detection against cluttered backgrounds, based on the MINACE (Minimum Noise and Correlation Energy) filter. Application of the filter is followed by a suitable segmentation algorithm, and the standard techniques of global and local thresholding are compared to watershed-based segmentation. The aim of this approach is to provide a custom region-based object detection algorithm with a concise set of regions of interest. Two industrial case studies are examined: diamond detection in X-ray images, and the reading of a dynamic, and ink stamped, 2D barcode on packaging clutter. We demonstrate the robustness of our approach on these two diverse applications, and develop a complete algorithmic prototype for an automatic stamped code reader. / AFRIKAANSE OPSOMMING: Hierdie tesis ondersoek die herkenning van voorwerpe teen onduidelike agtergronde. Ons benadering maak staat op die MINACE (" Minimum Noise and Correlation Energy") korrelasiefilter. Die filter word aangewend saam met 'n gepaste segmenteringsalgoritme, en die standaard tegnieke van globale en lokale drumpelingsalgoritmes word vergelyk met 'n waterskeidingsgebaseerde segmenteringsalgoritme. Die doel van hierdie deteksiebenadering is om 'n klein stel moontlike voorwerpe te kan verskaf aan enige klassifikasie-algoritme wat fokus op die voorwerpe self. Twee industriële toepassings word ondersoek: die opsporing van diamante in X-straal beelde, en die lees van 'n dinamiese, inkgedrukte, 2D balkieskode op verpakkingsmateriaal. Ons demonstreer die robuustheid van ons benadering met hierdie twee uiteenlopende voorbeelde, en ontwikkel 'n volledige algoritmiese prototipe vir 'n outomatiese stempelkode leser.

In this thesis, we study modellings and numerical algorithms to data analysis with applications to image processing and financial forecast. The thesis is composed of two parts, namely the tensor regression and data assimilation methods for image restoration.;We start with investigating the tensor regression problem in Chapter 2. It is a generalization of a classical regression in order to adopt and analyze much more information by using multi-dimensional arrays. Since the regression problem is subject to multiple solutions, we propose a regularized tensor regression model to the problem. By imposing a low-rank property of the solution and considering the structure of the tensor product, we develop an algorithm which is suitable for scalable implementations. The regularization method is used to select useful solutions which depend on applications. The proposed model is solved by the alternating minimization method and we prove the convergence of the objective function values and iterates by the maximization-minimization (MM) technique. We study different factors which affects the performance of the algorithm, including sample sizes, solution ranks and the noise levels. Applications include image compressing and financial forecast.;In Chapter 3, we apply filtering methods in data assimilation to image restoration problems. Traditionally, data assimilation methods optimally combine a predictive state from a dynamical system with real partially observations. The motivation is to improve the model forecast by real observation. We construct an artificial dynamics to the non-blind deblurring problems. By making use of spatial information of a single image, a span of ensemble members is constructed. A two-stage use of ensemble transform Kalman filter (ETKF) is adopted to deblur corrupted images. The theoretical background of ETKF and the use of artificial dynamics by stage augmentation method are provided. Numerical experiments include image and video processing.;Concluding remarks and discussion on future extensions are included in Chapter 4.

Chan, Man Leung Dunco. / Thesis submitted in: Nov 2008. / Thesis (M.Phil.)--Chinese University of Hong Kong, 2009. / Includes bibliographical references (leaves 87-91). / Abstracts in English and Chinese. / Abstract --- p.2 / 摘要 --- p.3 / Acknowledgement --- p.4 / Content --- p.5 / List of figures --- p.6 / Chapter Chapter 1 --- Introduction --- p.9 / Chapter 1.1 --- Background --- p.9 / Chapter 1.2 --- Prior work --- p.11 / Chapter 1.3 --- Objectives --- p.13 / Chapter 1.4 --- Proposed method --- p.16 / Chapter 1.5 --- Thesis outline --- p.18 / Chapter Chapter 2 --- Axial curve-pair deformation --- p.19 / Chapter 2.1 --- Axial deformation technique --- p.20 / Chapter 2.1.1 --- Representing objects in axial space --- p.21 / Chapter 2.1.2 --- Defining the frame --- p.23 / Chapter 2.2 --- Axial curve-pair deformation technique --- p.24 / Chapter 2.2.1 --- Framing the curve-pair --- p.25 / Chapter 2.2.2 --- Construction of orientation curve --- p.26 / Chapter 2.2.3 --- Manipulation of the axial curve-pair --- p.28 / Chapter Chapter 3 --- Dynamic axial curve-pair based deformation --- p.32 / Chapter 3.1 --- The dynamic mass spring model --- p.34 / Chapter 3.1.1 --- Dynamic NURBS curve --- p.35 / Chapter 3.1.2 --- Dynamic Free-form deformation --- p.37 / Chapter 3.1.3 --- Dynamic Axial Curve-pair deformation --- p.38 / Chapter 3.2 --- The dynamic mass spring model --- p.41 / Chapter 3.2.1 --- Curve-pair Fitting --- p.41 / Chapter 3.2.2 --- Construction of dynamic curve-pair --- p.44 / Chapter 3.2.3 --- The three-degree torsional spring --- p.48 / Chapter 3.2.4 --- Conserving feature in a twisting deformation --- p.50 / Chapter 3.2.5 --- Comparison of mass spring model --- p.51 / Chapter 3.3 --- Internal and external forces --- p.54 / Chapter 3.3.1 --- Tensile stress --- p.54 / Chapter 3.3.2 --- Torsional stress --- p.55 / Chapter 3.3.3 --- External forces --- p.59 / Chapter 3.4 --- Equations of motion --- p.60 / Chapter 3.5 --- System solver --- p.63 / Chapter 3.6 --- Hierarchical representation --- p.67 / Chapter 3.7 --- Collision detection --- p.72 / Chapter Chapter 4 --- Implementation and experimental result --- p.75 / Chapter 4.1 --- Comparison with original mass-spring system --- p.76 / Chapter 4.2 --- Comparison with dynamic free form deformation --- p.77 / Chapter 4.3 --- Comparison with the axial curve-pair deformation --- p.78 / Chapter 4.4 --- Shape restoring power --- p.80 / Chapter 4.5 --- Applications --- p.81 / Chapter Chapter 5 --- Conclusion --- p.84 / Reference --- p.86

Animating an articulated figure usually requires expensive hardware in terms of motion capture equipment, processing power and rendering power. This implies a high cost system and thus eliminates the use of personal computers to drive avatars in virtual environments. We propose a system to animate an articulated human upper body in real-time, using minimal motion capture trackers to provide position and orientation for the limbs. The system has to drive an avatar in a virtual environment on a low-end computer. The cost of the motion capture equipment must be relatively low (hence the use of minimal trackers). We discuss the various types of motion capture equipment and decide to use electromagnetic trackers which are adequate for our requirements while being reasonably priced. We also discuss the use of inverse kinematics to solve for the articulated chains making up the topology of the articulated figure. Furthermore, we offer a method to describe articulated chains as well as a process to specify the reach of up to four link chains with various levels of redundancy for use in articulated figures. We then provide various types of constraints to reduce the redundancy of non-defined articulated chains, specifically for chains found in an articulated human upper body. Such methods include a way to solve for the redundancy in the orientation of the neck link, as well as three different methods to solve the redundancy of the articulated human arm. The first method involves eliminating a degree of freedom from the chain, thus reducing its redundancy. The second method calculates the elevation angle of the elbow position from the elevation angle of the hand. The third method determines the actual position of the elbow from an average of previous positions of the elbow according to the position and orientation of the hand. The previous positions of the elbow are captured during the calibration process. The redundancy of the neck is easily solved due to the small amount of redundancy in the chain. When solving the arm, the first method which should give a perfect result in theory, gives a poor result in practice due to the limitations of both the motion capture equipment and the design. The second method provides an adequate result for the position of the redundant elbow in most cases although fails in some cases. Still it benefits from a simple approach as well as very little need for calibration. The third method provides the most accurate method of the three for the position of the redundant elbow although it also fails in some cases. This method however requires a long calibration session for each user. The last two methods allow for the calibration data to be used in latter session, thus reducing considerably the calibration required. In combination with a virtual reality system, these processes allow for the real-time animation of an articulated figure to drive avatars in virtual environments or for low quality animation on a low-end computer.

Thesis (Masters Diploma (Electrical Engineering)--Cape Technikon, Cape Town, 1993 / This thesis describes the design and development of a computer vision system. The system (Video Frame Grabber) will give PCUsers the potential to capture any visual image into the memory of a computer. This computer intelligible image opens the way for new development in computer photography, Image recognition and . Desktop Publishing.

Orientador: Renato da Rocha Lopes / Tese (doutorado) - Universidade Estadual de Campinas, Faculdade de Engenharia Elétrica e de Computação / Made available in DSpace on 2018-08-25T08:10:05Z (GMT). No. of bitstreams: 1 Takahata_AndreKazuo_D.pdf: 10418132 bytes, checksum: 71c4bb92534425059f1397eb4fc919a2 (MD5) Previous issue date: 2014 / Resumo: Neste trabalho consideramos técnicas de processamento de sinais que têm como objetivo aumentar a resolução de imagens da subsuperfície geradas com dados sísmicos. Uma das técnicas consideradas é a deconvolução unidimensional, que tem como finalidade eliminar distorções causadas pelas limitações em banda de frequência da fonte sísmica, bem como pela absorção de componentes e distorções de fase ocorridas durante a propagação da onda sísmica. Nesta tese, analisamos tanto métodos chamados supervisionados, em que estão disponíveis medidas adicionais às medidas sísmicas, que podem guiar o processo de deconvolução, quanto os métodos não supervisionados, em que apenas as medidas sísmicas são consideradas. Em particular, tratamos dos métodos de filtragem de Wiener e mínimos quadrados para os métodos supervisionados. Nos métodos não supervisionados, discutimos as hipóteses para o funcionamento dos métodos envolvendo as estatísticas referentes à refletividade de subsuperfície e do espectro de fase do pulso sísmico. Em particular, analisamos principalmente uso do filtro de erro de predição, que utiliza estatísticas de segunda ordem (SOS) e requer um pulso de fase mínima, e mostramos nossa contribuição sobre um método que utiliza estatísticas de ordem superior (HOS) chamado de "banded independent component analysis" (B-ICA) e que não exige que o pulso seja de fase mínima. Por fim, realizamos um estudo de caso envolvendo dados obtidos em um poço e dados sísmicos com fim de ilustrar nossa análise. Na deconvolução bidimensional são tratadas, além das distorções pela fonte sísmica consideradas na abordagem unidimensional, distorções causadas pela geometria de aquisição de dados e de variações de velocidade de propagação sísmica causadas por complexidades geológicas. Tais distorções podem ser quantificadas em imagens sísmicas obtidas pela técnica de imageamento chamada migração em profundidade pré-empilhamento (PSDM) por meio de uma relação de convolução bidimensional entre a refletividade da subsuperfície e uma função de resolução. Sob hipóteses adequadas, a função de resolução pode ser modelada como uma função de espalhamento pontual (PSF) e a deconvolução bidimensional, portanto, consiste em atenuar o efeito dessas PSFs. Neste trabalho revisamos os aspectos básicos desta modelagem e da estimação das PSFs, bem como do processo de imageamento, e mostramos a nossa contribuição para a deconvolução bidimensional por meio de um método de filtragem inversa / Abstract: In this work, we consider signal processing techniques that aim to improve the resolution of images of the subsurface of the Earth generated from seismic data. One such technique is uni-dimensional deconvolution, which aims to eliminate distortions caused by limitations in the seismic source frequency band, as well as distorting effects caused by frequency components absorption and phase changes during seismic propagation. We analyze both supervised methods, in which reference signals are used in addition to the seismic measurements to determine the decovolution filter, as well as unsupervised methods, in which only the seismic measurements are used. Particularly, we analyze Wiener filtering and least squares methods on the supervised case. As for the unsupervised algorithms, we discuss the hypotheses that underlie these methods, which are based on the statistics of the reflectivity of the subsurface and the phase spectrum of the wavelet pulse. We analyze especially the use of the prediction error filter, which uses second order statistics (SOS) and requires a minimum phase wavelet, and we show our contribution on a method that uses higher order statistics (HOS) called banded independent component analysis (B-ICA), which does not requires that the wavelet be minimum phase. We also present a case study using log data measured in a borehole and seismic data in order to illustrate our analysis. In bidimensional deconvolution, we consider, besides the seismic source distortions considered in the 1D approach, distortions in seismic imaging caused by the acquisition geometry and velocity model complexity associated with the geological structure of the subsurface. These distortions can be quantified in seismic images created through the technique called prestack depth migration (PSDM) using a 2D convolution model between the reflectivity of the subsurface and the so-called resolution function. Under appropriate hypotheses, the resolution function can be seen as a point spread function (PSF). Thus, the objective of 2D deconvolution is to attenuate the effect of these PSFs. In this work, we review the basic aspects of the 2D convolutional model and PSF estimation, as well as the imaging process, and we show our contribution on 2D deconvolution using an inverse filtering approach / Doutorado / Telecomunicações e Telemática / Doutor em Engenharia Elétrica

This study sought to determine whether or not it may be possible to evoke ‘the musical effect' – the emotional response perceived by music listeners – using white noise as a sound-source and real-time digital signal processing techniques. This information was considered to be valuable as in a world driven by technological progress the potential use of new or different technologies in creating music could lead to the development of new methods of – and tools for – composition and performance. More specifically this research asked the question 'what is music?' and investigated how humans – both trained musicians and untrained people – perceive it. The elements of music were investigated for their affective strengths and new fields of research explored for insights into emotion identification in music. Thereafter the focus shifted into the realm of Digital Signal Processing. Common operations and techniques for signal manipulation were investigated and an understanding of the field as a whole was sought. The culmination of these two separate, yet related, investigations was the design and implementation of a listening experiment conducted on adult subjects. They were asked to listen to various manipulated noise-signals and answer a questionnaire with regard to their perceptions of the audio material. The data from the listening experiment suggest that certain DSP techniques can evoke ‘the musical effect’. Various musical elements were represented via digital techniques and in many cases respondents reported perceptions which suggest that some effect was felt. The techniques implemented and musical elements represented were discussed, and possible applications for these techniques, both musical and non-musical, were explored. Areas for further research were discussed and include the implementation of even more DSP techniques, and also into garnering a more specific idea of the emotion perceived by respondents in response to the experiment material.

The subject of this dissertation is coherent digital acoustic communication in the underwater environment. The objective of the research is to develop algorithms for reliable communication in the shallow underwater channel. Investigation is focused on channel depth less than 100 m and distances between transmitter and receiver from 5 km to 50 km. Based on the characteristics of the underwater acoustic channel and using a conventional approach, the achievable transmission range and the required acoustic power is determined for given channel conditions and system parameters. A channel model suitable for the investigation of shallow water communication is developed which takes into account transmitter-receiver geometry, environmental conditions and system parameters. The model is based on multiple reflections in the channel with weightings according to signal attenuation due to spreading, reflection losses and absorption. Time-variability is introduced by incorporating Doppler frequency shifts due to transmitter/receiver motion. A new method of evaluating performance of a system operating in such multipath conditions is proposed by introduction of a signal-to-multipath ratio (SMR), which is a measure of intersymbol interference (ISI) caused by the multipath. The SMR allows assessment of system performance for various receiver/transmitter positions and channel parameters. It can be used, for instance, to find the transmitter/receiver depth for optimum transmission. A suitable equalizer can improve a SMR. For example, a decision feedback equalizer (DFE) using a least mean square (LMS) and fast optimized LMS criterion is effective in coping with ISI as demonstrated by computer simulations. Hardware complexities of several equalizer algorithms are investigated for a selected channel. The performance degradation due to the presence of Gaussian noise in addition to multipath is analyzed by simulation. A novel structure of an equalizer suitable for the time-variant underwater acoustic channel is proposed. By adaptively adjusting the number of equalizer taps depending on the channel condition, the proposed structure offers reduced hardware complexity. Computer simulations demonstrate the effectiveness of this approach. It is anticipated that the results of this work will find application in the design of high data rate transmission systems for ocean bottom instrumentation, the design of telemetry for autonomous underwater vehicles, and others. / Graduate

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