Yes / n this paper, a powerful face recognition system for authentication and identification tasks is presented and a new facial feature extraction approach is proposed. A novel feature extraction method based on combining the characteristics of the Curvelet transform and Fractal dimension transform is proposed. The proposed system consists of four stages. Firstly, a simple preprocessing algorithm based on a sigmoid function is applied to standardize the intensity dynamic range in the input image. Secondly, a face detection stage based on the Viola-Jones algorithm is used for detecting the face region in the input image. After that, the feature extraction stage using a combination of the Digital Curvelet via wrapping transform and a Fractal Dimension transform is implemented. Finally, the K-Nearest Neighbor (K-NN) and Correlation Coefficient (CC) Classifiers are used in the recognition task. Lastly, the performance of the proposed approach has been tested by carrying out a number of experiments on three well-known datasets with high diversity in the facial expressions: SDUMLA-HMT, Faces96 and UMIST datasets. All the experiments conducted indicate the robustness and the effectiveness of the proposed approach for both authentication and identification tasks compared to other established approaches.

In this thesis examples of spectral triples, which represent fractal sets, are examined and new insights into their noncommutative geometries are obtained. Firstly, starting with Connes' spectral triple for a non-empty compact totally disconnected subset E of {R} with no isolated points, we develop a noncommutative coarse multifractal formalism. Specifically, we show how multifractal properties of a measure supported on E can be expressed in terms of a spectral triple and the Dixmier trace of certain operators. If E satisfies a given porosity condition, then we prove that the coarse multifractal box-counting dimension can be recovered. We show that for a self-similar measure μ, given by an iterated function system S defined on a compact subset of {R} satisfying the strong separation condition, our noncommutative coarse multifractal formalism gives rise to a noncommutative integral which recovers the self-similar multifractal measure ν associated to μ, and we establish a relationship between the noncommutative volume of such a noncommutative integral and the measure theoretical entropy of ν with respect to S. Secondly, motivated by the results of Antonescu-Ivan and Christensen, we construct a family of (1, +)-summable spectral triples for a one-sided topologically exact subshift of finite type (∑{{A}} {{N}}, σ). These spectral triples are constructed using equilibrium measures obtained from the Perron-Frobenius-Ruelle operator, whose potential function is non-arithemetic and Hölder continuous. We show that the Connes' pseudo-metric, given by any one of these spectral triples, is a metric and that the metric topology agrees with the weak*-topology on the state space {S}(C(∑{{A}} {{N}}); {C}). For each equilibrium measure ν[subscript(φ)] we show that the noncommuative volume of the associated spectral triple is equal to the reciprocal of the measure theoretical entropy of ν[subscript(φ)] with respect to the left shift σ (where it is assumed, without loss of generality, that the pressure of the potential function is equal to zero). We also show that the measure ν[subscript(φ)] can be fully recovered from the noncommutative integration theory.

This thesis concerns an active research area within fractal geometry. In the first part, in Chapters 2 and 3, for directed graph iterated function systems (IFSs) defined on ℝ, we prove that a class of 2-vertex directed graph IFSs have attractors that cannot be the attractors of standard (1-vertex directed graph) IFSs, with or without separation conditions. We also calculate their exact Hausdorff measure. Thus we are able to identify a new class of attractors for which the exact Hausdorff measure is known. We give a constructive algorithm for calculating the set of gap lengths of any attractor as a finite union of cosets of finitely generated semigroups of positive real numbers. The generators of these semigroups are contracting similarity ratios of simple cycles in the directed graph. The algorithm works for any IFS defined on ℝ with no limit on the number of vertices in the directed graph, provided a separation condition holds. The second part, in Chapter 4, applies to directed graph IFSs defined on ℝⁿ . We obtain an explicit calculable value for the power law behaviour as r → 0⁺ , of the qth packing moment of μ[subscript(u)], the self-similar measure at a vertex u, for the non-lattice case, with a corresponding limit for the lattice case. We do this (i) for any q ∈ ℝ if the strong separation condition holds, (ii) for q ≥ 0 if the weaker open set condition holds and a specified non-negative matrix associated with the system is irreducible. In the non-lattice case this enables the rate of convergence of the packing L[superscript(q)]-spectrum of μ[subscript(u)] to be determined. We also show, for (ii) but allowing q ∈ ℝ, that the upper multifractal q box-dimension with respect to μ[subscript(u)], of the set consisting of all the intersections of the components of F[subscript(u)], is strictly less than the multifractal q Hausdorff dimension with respect to μ[subscript(u)] of F[subscript(u)].

Efficient temporal resolution and spatial grids are important in simulation of the inflow turbulence for wind turbine loads analyses. There have not been many published studies that address optimal space-time resolution of generated inflow velocity fields in order to estimate accurate load statistics. This study investigates turbine extreme and fatigue load statistics for a utility-scale 5MW wind turbine with a hub-height of 90 m and a rotor diameter of 126 m. Load statistics, spectra, and time-frequency analysis representations are compared for various alternative space and time resolutions employed in inflow turbulence field simulation. Conclusions are drawn regarding adequate resolution in space of the inflow turbulence simulated on the rotor plane prior to extracting turbine load statistics. Similarly, conclusions are drawn with regard to what constitutes adequate temporal filtering to preserve turbine load statistics. This first study employs conventional Fourier-based spectral methods for stochastic simulation of velocity fields for a neutral atmospheric boundary layer. In the second part of this study, large-eddy simulation (LES) is employed with similar resolutions in space and time as in the earlier Fourier-based simulations to again establish turbine load statistics. A comparison of extreme and fatigue load statistics is presented for the two approaches used for inflow field generation. The use of LES-generated flows (enhanced in deficient high-frequency energy by the use of fractal interpolation) to establish turbine load statistics in this manner is computationally very expensive but the study is justified in order to evaluate the ability of LES to be used as an alternative to more common approaches. LES with fractal interpolation is shown to lead to accurate load statistics when compared with stochastic simulation. A more compelling reason for using LES in turbine load studies is the following: for stable boundary layers, it is not possible to generate realistic inflow velocity fields using stochastic simulation. The present study presents a demonstration that, despite the computational costs involved, LES-generated inflows can be used for loads analyses for utility-scale turbines. The study sets the stage for future computations in the stable boundary layer where low-level jets, large speed and direction shears across the rotor, etc. can possibly cause large turbine loads; then, LES will likely be the inflow turbulence generator of choice. / text

L'objet de cette thèse est l'analyse multifractale des fonctions autosimilaires et l'étude de la validité du formalisme multifractal. Il s'agit d'abord de déterminer la régularité Hölderienne ponctuelle exacte pour des fonctions dont le graphe localement est grossièrement une contraction du graphe complet, à une fonction erreur près ; ensuite de calculer les dimensions de Hausdorff des ensembles de points où la fonction présente la même singularité; et enfin de vérifier les conjectures de Frish et Parisi et celle d'Arneodo, Bacry et Muzy, qui relient ces dimensions à des quantités moyennes extraites de la fonction. Nous étudions plusieurs types d'autosimilarités, et montrons (en reformulant parfois) que l'analyse par ondelettes permet d'étudier la validité de ces relations.

Mémoire numérisé par la Division de la gestion de documents et des archives de l'Université de Montréal.

In everyday situations individuals make decisions. For example, a tourist usually chooses a crowded or recommended restaurant to have dinner. Perhaps it is an individual decision, but the observed pattern of decision-making is a collective phenomenon. Collective behaviour emerges from the local interactions that give rise to a complex pattern at the group level. In our example, the recommendations or simple copying the choices of others make a crowded restaurant even more crowded. The rules of interaction between individuals are important to study. Such studies should be complemented by biological experiments. Recent studies of collective phenomena in animal groups help us to understand these rules and develop mathematical models of collective behaviour. The most important communication mechanism is positive feedback between group members, which we observe in our example. In this thesis, we use a generic experimentally validated model of positive feedback to study collective decision-making. The first part of the thesis is based on the modelling of decision-making associated to the selection of feeding sites. This has been extensively studied for ants and slime moulds. The main contribution of our research is to demonstrate how such aspects as "irrationality", speed and quality of decisions can be modelled using differential equations. We study bifurcation phenomena and describe collective patterns above critical values of a bifurcation points in mathematical and biological terms. In the second part, we demonstrate how the primitive unicellular slime mould Physarum Polycephalum provides an easy test-bed for theoretical assumptions and model predictions about decision-making. We study its searching strategies and model decision-making associated to the selection of food options. We also consider the aggregation model to investigate the fractal structure of Physarum Polycephalum plasmodia. / <p>Fel serie i tryckt bok /Wrong series in the printed book</p>

Given a very large dataset of moderate-to-high dimensionality, how to mine useful patterns from it? In such cases, dimensionality reduction is essential to overcome the well-known curse of dimensionality. Although there exist algorithms to reduce the dimensionality of Big Data, unfortunately, they all fail to identify/eliminate non-linear correlations that may occur between the attributes. This MSc work tackles the problem by exploring concepts of the Fractal Theory and massive parallel processing to present Curl-Remover, a novel dimensionality reduction technique for very large datasets. Our contributions are: (a) Curl-Remover eliminates linear and non-linear attribute correlations as well as irrelevant attributes; (b) it is unsupervised and suits for analytical tasks in general not only classification; (c) it presents linear scale-up on both the data size and the number of machines used; (d) it does not require the user to guess the number of attributes to be removed, and; (e) it preserves the attributes semantics by performing feature selection, not feature extraction. We executed experiments on synthetic and real data spanning up to 1.1 billion points, and report that our proposed Curl-Remover outperformed two PCA-based algorithms from the state-of-the-art, being in average up to 8% more accurate. / Dada uma grande base de dados de dimensionalidade moderada a alta, como identificar padrões úteis nos objetos de dados? Nesses casos, a redução de dimensionalidade é essencial para superar um fenômeno conhecido na literatura como a maldição da alta dimensionalidade. Embora existam algoritmos capazes de reduzir a dimensionalidade de conjuntos de dados na escala de Terabytes, infelizmente, todos falham em relação à identificação/eliminação de correlações não lineares entre os atributos. Este trabalho de Mestrado trata o problema explorando conceitos da Teoria de Fractais e processamento paralelo em massa para apresentar Curl-Remover, uma nova técnica de redução de dimensionalidade bem adequada ao pré-processamento de Big Data. Suas principais contribuições são: (a) Curl-Remover elimina correlações lineares e não lineares entre atributos, bem como atributos irrelevantes; (b) não depende de supervisão do usuário e é útil para tarefas analíticas em geral não apenas para a classificação; (c) apresenta escalabilidade linear tanto em relação ao número de objetos de dados quanto ao número de máquinas utilizadas; (d) não requer que o usuário sugira um número de atributos para serem removidos, e; (e) mantêm a semântica dos atributos por ser uma técnica de seleção de atributos, não de extração de atributos. Experimentos foram executados em conjuntos de dados sintéticos e reais contendo até 1,1 bilhões de pontos, e a nova técnica Curl-Remover apresentou desempenho superior comparada a dois algoritmos do estado da arte baseados em PCA, obtendo em média até 8% a mais em acurácia de resultados.

Estudos mostraram que o tráfego nas redes de dados tanto locais quanto de grande área, possui propriedades fractais como dependência de longa duração - Long-Range Dependence (LRD) e auto-similaridade. Devido à heterogeneidade de aplicações nessas redes, os traces de tráfego podem apresentar dependência de longa duração - Long Range Dependence (LRD), dependência de curta duração - Short Range Dependence (SRD) ou uma mistura de LRD com SRD. Sendo assim, este trabalho tem como objetivo sintetizar séries temporais gaussianas com flexibilidade de processamento no plano tempo-frequência a serem inseridas num gerador de tráfego com as características estatísticas específicas do tráfego encontrado em redes por comutação de pacotes reais, como autossimilaridade, LRD e SRD. Para isto foram desenvolvidos dois métodos para síntese de séries temporais gaussianas com LRD e simultânea introdução de SRD em diferentes faixas de frequência: Discrete Wavelet Tansform (DWT) com mapa de variâncias e Discrete Wavelet Packet Tansform (DWPT). Estes métodos utilizaram o mapa de variâncias cujo conceito foi desenvolvido neste trabalho. A validação dos métodos foi feita através de análise estatística e comparação com resultados de séries geradas pelo método Discrete Wavelet Transfom (DWT) de Backar utilizado em [1]. Além disso, também foi validada a ideia de que a DWPT é mais interessante que a DWT por ser mais flexível e prover uma maior flexibilidade de processamento no plano tempo-frequência. / Studies demonstrated that the data network traffic of Local Area Network (LAN) and Wide Area Network has fractal properties as long range dependence (LRD) and self-similarity. The traffic traces can show long range dependence, short range dependence or the both behaviors because of applications heterogeneity in these networks. This work objective is to synthetisize gaussian time series with processor flexibility in the time-frequency plan to be inserted in a traffic generator with the specific statistical traffic characteristics of real packet networks such as selfsimilarity, long range dependence (LRD) and short range dependence (SRD). Two methods were developed for the gaussian time series with LRD and SRD synthesis: Discrete Wavelet Tansform (DWT) with variance map and Discrete Wavelet Packet Tansform (DWPT). These methods used the variance map which concept was developed in this work. The methods validation was done by statistic analysis and comparison with the time series generated by the B¨ackar Discrete Wavelet Transfom (DWT) used by [1]. Besides of this, the idea that the DWPT is more because of its processing flexibility in the time-frequency plan was validated.

Occlusion is a procedurally randomized interactive art experience which uses the motifs of repetition, isolation, incongruity and mutability to develop an experience of a Folie àDeux: a madness shared by two. It draws from traditional video game forms, development methods, and tools to situate itself in context with games as well as other forms of interactive digital media. In this way, Occlusion approaches the making of game-like media from the art criticism perspective of Materiality, and the written work accompanying the prototype discusses critical aesthetic concerns for Occlusion both as an art experience borrowing from games and as a text that can be academically understood in relation to other practices of media making. In addition to the produced software artifact and written analysis, this thesis includes primary research in the form of four interviews with artists, authors, game makers and game critics concerning Materiality and dissociative themes in game-like media. The written work first introduces Occlusion in context with other approaches to procedural remixing, Glitch Art, net.art, and analogue and digital collage and décollage, with special attention to recontextualization and apophenia. The experience, visual, and audio design approach of Occlusion is reviewed through a discussion of explicit design choices which define generative space. Development process, release process, post-release distribution, testing, and maintenance are reviewed, and the paper concludes with a description of future work and a post- mortem discussion. Included as appendices are a full specification document, script, and transcripts of all interviews.