Image enhancement with MatLab algorithms

Nowadays, a lot of applications use digital images. For example in face recognition to detect and tag persons in photograph, for security control, and a lot of applications that can be found in smart cities, as speed control in roads or highways and cameras in traffic lights to detect drivers ignoring red light. Also in medicine digital images are used, such as x-ray, scanners, etc. These applications depend on the quality of the image obtained. A good camera is expensive, and the image obtained depends also on external factor as light. To make these applications work properly, image enhancement is as important as, for example, a good face detection algorithm. Image enhancement also can be used in normal photograph, for pictures done in bad light conditions, or just to improve the contrast of an image. There are some applications for smartphones that allow users apply filters or change the bright, colour or contrast on the pictures. This project compares four different techniques to use in image enhancement. After applying one of these techniques to an image, it will use better the whole available dynamic range. Some of the algorithms are designed for grey scale images and others for colour images. It is used Matlab software to develop and present the final results. These algorithms are Successive Means Quantization Transform (SMQT), Histogram Equalization, using Matlab function and own implemented function, and V transform. Finally, as conclusions, we can prove that Histogram equalization algorithm is the simplest of all, it has a wide variability of grey levels and it is not suitable for colour images. V transform algorithm is a good option for colour images. The algorithm is linear and requires low computational power. SMQT algorithm is non-linear, insensitive to gain and bias and it can extract structure of the data. RESUMEN. Hoy en día incontable número de aplicaciones usan imágenes digitales. Por ejemplo, para el control de la seguridad se usa el reconocimiento de rostros para detectar y etiquetar personas en fotografías o vídeos, para distintos usos de las ciudades inteligentes, como control de velocidad en carreteras o autopistas, cámaras en los semáforos para detectar a conductores haciendo caso omiso de un semáforo en rojo, etc. También en la medicina se utilizan imágenes digitales, como por ejemplo, rayos X, escáneres, etc. Todas estas aplicaciones dependen de la calidad de la imagen obtenida. Una buena cámara es cara, y la imagen obtenida depende también de factores externos como la luz. Para hacer que estas aplicaciones funciones correctamente, el tratamiento de imagen es tan importante como, por ejemplo, un buen algoritmo de detección de rostros. La mejora de la imagen también se puede utilizar en la fotografía no profesional o de consumo, para las fotos realizadas en malas condiciones de luz, o simplemente para mejorar el contraste de una imagen. Existen aplicaciones para teléfonos móviles que permiten a los usuarios aplicar filtros y cambiar el brillo, el color o el contraste en las imágenes. Este proyecto compara cuatro técnicas diferentes para utilizar el tratamiento de imagen. Se utiliza la herramienta de software matemático Matlab para desarrollar y presentar los resultados finales. Estos algoritmos son Successive Means Quantization Transform (SMQT), Ecualización del histograma, usando la propia función de Matlab y una nueva función que se desarrolla en este proyecto y, por último, una función de transformada V. Finalmente, como conclusión, podemos comprobar que el algoritmo de Ecualización del histograma es el más simple de todos, tiene una amplia variabilidad de niveles de gris y no es adecuado para imágenes en color. El algoritmo de transformada V es una buena opción para imágenes en color, es lineal y requiere baja potencia de cálculo. El algoritmo SMQT no es lineal, insensible a la ganancia y polarización y, gracias a él, se puede extraer la estructura de los datos.

Parallel Computer Vision Algorithms for Graphics Processing Units

La evolución de los teléfonos móviles inteligentes, dotados de cámaras digitales, está provocando una creciente demanda de aplicaciones cada vez más complejas que necesitan algoritmos de visión artificial en tiempo real; puesto que el tamaño de las señales de vídeo no hace sino aumentar y en cambio el rendimiento de los procesadores de un solo núcleo se ha estancado, los nuevos algoritmos que se diseñen para visión artificial han de ser paralelos para poder ejecutarse en múltiples procesadores y ser computacionalmente escalables. Una de las clases de procesadores más interesantes en la actualidad se encuentra en las tarjetas gráficas (GPU), que son dispositivos que ofrecen un alto grado de paralelismo, un excelente rendimiento numérico y una creciente versatilidad, lo que los hace interesantes para llevar a cabo computación científica. En esta tesis se exploran dos aplicaciones de visión artificial que revisten una gran complejidad computacional y no pueden ser ejecutadas en tiempo real empleando procesadores tradicionales. En cambio, como se demuestra en esta tesis, la paralelización de las distintas subtareas y su implementación sobre una GPU arrojan los resultados deseados de ejecución con tasas de refresco interactivas. Asimismo, se propone una técnica para la evaluación rápida de funciones de complejidad arbitraria especialmente indicada para su uso en una GPU. En primer lugar se estudia la aplicación de técnicas de síntesis de imágenes virtuales a partir de únicamente dos cámaras lejanas y no paralelas—en contraste con la configuración habitual en TV 3D de cámaras cercanas y paralelas—con información de color y profundidad. Empleando filtros de mediana modificados para la elaboración de un mapa de profundidad virtual y proyecciones inversas, se comprueba que estas técnicas son adecuadas para una libre elección del punto de vista. Además, se demuestra que la codificación de la información de profundidad con respecto a un sistema de referencia global es sumamente perjudicial y debería ser evitada. Por otro lado se propone un sistema de detección de objetos móviles basado en técnicas de estimación de densidad con funciones locales. Este tipo de técnicas es muy adecuada para el modelado de escenas complejas con fondos multimodales, pero ha recibido poco uso debido a su gran complejidad computacional. El sistema propuesto, implementado en tiempo real sobre una GPU, incluye propuestas para la estimación dinámica de los anchos de banda de las funciones locales, actualización selectiva del modelo de fondo, actualización de la posición de las muestras de referencia del modelo de primer plano empleando un filtro de partículas multirregión y selección automática de regiones de interés para reducir el coste computacional. Los resultados, evaluados sobre diversas bases de datos y comparados con otros algoritmos del estado del arte, demuestran la gran versatilidad y calidad de la propuesta. Finalmente se propone un método para la aproximación de funciones arbitrarias empleando funciones continuas lineales a tramos, especialmente indicada para su implementación en una GPU mediante el uso de las unidades de filtraje de texturas, normalmente no utilizadas para cómputo numérico. La propuesta incluye un riguroso análisis matemático del error cometido en la aproximación en función del número de muestras empleadas, así como un método para la obtención de una partición cuasióptima del dominio de la función para minimizar el error. ABSTRACT The evolution of smartphones, all equipped with digital cameras, is driving a growing demand for ever more complex applications that need to rely on real-time computer vision algorithms. However, video signals are only increasing in size, whereas the performance of single-core processors has somewhat stagnated in the past few years. Consequently, new computer vision algorithms will need to be parallel to run on multiple processors and be computationally scalable. One of the most promising classes of processors nowadays can be found in graphics processing units (GPU). These are devices offering a high parallelism degree, excellent numerical performance and increasing versatility, which makes them interesting to run scientific computations. In this thesis, we explore two computer vision applications with a high computational complexity that precludes them from running in real time on traditional uniprocessors. However, we show that by parallelizing subtasks and implementing them on a GPU, both applications attain their goals of running at interactive frame rates. In addition, we propose a technique for fast evaluation of arbitrarily complex functions, specially designed for GPU implementation. First, we explore the application of depth-image–based rendering techniques to the unusual configuration of two convergent, wide baseline cameras, in contrast to the usual configuration used in 3D TV, which are narrow baseline, parallel cameras. By using a backward mapping approach with a depth inpainting scheme based on median filters, we show that these techniques are adequate for free viewpoint video applications. In addition, we show that referring depth information to a global reference system is ill-advised and should be avoided. Then, we propose a background subtraction system based on kernel density estimation techniques. These techniques are very adequate for modelling complex scenes featuring multimodal backgrounds, but have not been so popular due to their huge computational and memory complexity. The proposed system, implemented in real time on a GPU, features novel proposals for dynamic kernel bandwidth estimation for the background model, selective update of the background model, update of the position of reference samples of the foreground model using a multi-region particle filter, and automatic selection of regions of interest to reduce computational cost. The results, evaluated on several databases and compared to other state-of-the-art algorithms, demonstrate the high quality and versatility of our proposal. Finally, we propose a general method for the approximation of arbitrarily complex functions using continuous piecewise linear functions, specially formulated for GPU implementation by leveraging their texture filtering units, normally unused for numerical computation. Our proposal features a rigorous mathematical analysis of the approximation error in function of the number of samples, as well as a method to obtain a suboptimal partition of the domain of the function to minimize approximation error.

Odor space and olfactory processing: Collective algorithms and neural implementation

Several basic olfactory tasks must be solved by highly olfactory animals, including background suppression, multiple object separation, mixture separation, and source identification. The large number N of classes of olfactory receptor cells—hundreds or thousands—permits the use of computational strategies and algorithms that would not be effective in a stimulus space of low dimension. A model of the patterns of olfactory receptor responses, based on the broad distribution of olfactory thresholds, is constructed. Representing one odor from the viewpoint of another then allows a common description of the most important basic problems and shows how to solve them when N is large. One possible biological implementation of these algorithms uses action potential timing and adaptation as the “hardware” features that are responsible for effective neural computation.

Dimensionality Reduction Through Sub-Space Mapping for Nearest Neighbour Algorithms

Postprint

Approximation algorithms

Increasing global competition, rapidly changing markets, and greater consumer awareness have altered the way in which corporations do business. To become more efficient, many industries have sought to model some operational aspects by gigantic optimization problems. It is not atypical to encounter models that capture 106 separate “yes” or “no” decisions to be made. Although one could, in principle, try all 2106 possible solutions to find the optimal one, such a method would be impractically slow. Unfortunately, for most of these models, no algorithms are known that find optimal solutions with reasonable computation times. Typically, industry must rely on solutions of unguaranteed quality that are constructed in an ad hoc manner. Fortunately, for some of these models there are good approximation algorithms: algorithms that produce solutions quickly that are provably close to optimal. Over the past 6 years, there has been a sequence of major breakthroughs in our understanding of the design of approximation algorithms and of limits to obtaining such performance guarantees; this area has been one of the most flourishing areas of discrete mathematics and theoretical computer science.

An iterative method for extracting energy-like quantities from protein structures.

We present a method (ENERGI) for extracting energy-like quantities from a data base of protein structures. In this paper, we use the method to generate pairwise additive amino acid "energy" scores. These scores are obtained by iteration until they correctly discriminate a set of known protein folds from decoy conformations. The method succeeds in lattice model tests and in the gapless threading problem as defined by Maiorov and Crippen [Maiorov, V. N. & Crippen, G. M. (1992) J. Mol. Biol. 227, 876-888]. A more challenging test of threading a larger set of test proteins derived from the representative set of Hobohm and Sander [Hobohm, U. & Sander, C. (1994) Protein Sci. 3, 522-524] is used as a "workbench" for exploring how the ENERGI scores depend on their parameter sets.

Chaperonin-facilitated protein folding: optimization of rate and yield by an iterative annealing mechanism.

We develop a heuristic model for chaperonin-facilitated protein folding, the iterative annealing mechanism, based on theoretical descriptions of "rugged" conformational free energy landscapes for protein folding, and on experimental evidence that (i) folding proceeds by a nucleation mechanism whereby correct and incorrect nucleation lead to fast and slow folding kinetics, respectively, and (ii) chaperonins optimize the rate and yield of protein folding by an active ATP-dependent process. The chaperonins GroEL and GroES catalyze the folding of ribulose bisphosphate carboxylase at a rate proportional to the GroEL concentration. Kinetically trapped folding-incompetent conformers of ribulose bisphosphate carboxylase are converted to the native state in a reaction involving multiple rounds of quantized ATP hydrolysis by GroEL. We propose that chaperonins optimize protein folding by an iterative annealing mechanism; they repeatedly bind kinetically trapped conformers, randomly disrupt their structure, and release them in less folded states, allowing substrate proteins multiple opportunities to find pathways leading to the most thermodynamically stable state. By this mechanism, chaperonins greatly expand the range of environmental conditions in which folding to the native state is possible. We suggest that the development of this device for optimizing protein folding was an early and significant evolutionary event.

Interaction of GroEL with a highly structured folding intermediate: iterative binding cycles do not involve unfolding.

The GroE proteins are molecular chaperones involved in protein folding. The general mechanism by which they facilitate folding is still enigmatic. One of the central open questions is the conformation of the GroEL-bound nonnative protein. Several suggestions have been made concerning the folding stage at which a protein can interact with GroEL. Furthermore, the possibility exists that binding of the nonnative protein to GroEL results in its unfolding. We have addressed these issues that are basic for understanding the GroE-mediated folding cycle by using folding intermediates of an Fab antibody fragment as molecular probes to define the binding properties of GroEL. We show that, in addition to binding to an early folding intermediate, GroEL is able to recognize and interact with a late quaternary-structured folding intermediate (Dc) without measurably unfolding it. Thus, the prerequisite for binding is not a certain folding stage of a nonnative protein. In contrast, general surface properties of nonnative proteins seem to be crucial for binding. Furthermore, unfolding of a highly structured intermediate does not necessarily occur upon binding to GroEL. Folding of Dc in the presence of GroEL and ATP involves cycles of binding and release. Because in this system no off-pathway reactions or kinetic traps are involved, a quantitative analysis of the reactivation kinetics observed is possible. Our results indicate that the association reaction of Dc and GroEL in the presence of ATP is rather slow, whereas in the absence of ATP association is several orders of magnitude more efficient. Therefore, it seems that ATP functions by inhibiting reassociation rather than promoting release of the bound substrate.

Método multigrid algébrico: reutilização das estruturas multigrid no transporte de contaminantes

A necessidade de obter solução de grandes sistemas lineares resultantes de processos de discretização de equações diferenciais parciais provenientes da modelagem de diferentes fenômenos físicos conduz à busca de técnicas numéricas escaláveis. Métodos multigrid são classificados como algoritmos escaláveis.Um estimador de erros deve estar associado à solução numérica do problema discreto de modo a propiciar a adequada avaliação da solução obtida pelo processo de aproximação. Nesse contexto, a presente tese caracteriza-se pela proposta de reutilização das estruturas matriciais hierárquicas de operadores de transferência e restrição dos métodos multigrid algébricos para acelerar o tempo de solução dos sistemas lineares associados à equação do transporte de contaminantes em meio poroso saturado. Adicionalmente, caracteriza-se pela implementação das estimativas residuais para os problemas que envolvem dados constantes ou não constantes, os regimes de pequena ou grande advecção e pela proposta de utilização das estimativas residuais associadas ao termo de fonte e à condição inicial para construir procedimentos adaptativos para os dados do problema. O desenvolvimento dos códigos do método de elementos finitos, do estimador residual e dos procedimentos adaptativos foram baseados no projeto FEniCS, utilizando a linguagem de programação PYTHONR e desenvolvidos na plataforma Eclipse. A implementação dos métodos multigrid algébricos com reutilização considera a biblioteca PyAMG. Baseado na reutilização das estruturas hierárquicas, os métodos multigrid com reutilização com parâmetro fixo e automática são propostos, e esses conceitos são estendidos para os métodos iterativos não-estacionários tais como GMRES e BICGSTAB. Os resultados numéricos mostraram que o estimador residual captura o comportamento do erro real da solução numérica, e fornece algoritmos adaptativos para os dados cuja malha retornada produz uma solução numérica similar à uma malha uniforme com mais elementos. Adicionalmente, os métodos com reutilização são mais rápidos que os métodos que não empregam o processo de reutilização de estruturas. Além disso, a eficiência dos métodos com reutilização também pode ser observada na solução do problema auxiliar, o qual é necessário para obtenção das estimativas residuais para o regime de grande advecção. Esses resultados englobam tanto os métodos multigrid algébricos do tipo SA quanto os métodos pré-condicionados por métodos multigrid algébrico SA, e envolvem o transporte de contaminantes em regime de pequena e grande advecção, malhas estruturadas e não estruturadas, problemas bidimensionais, problemas tridimensionais e domínios com diferentes escalas.

High speed imaging and algorithms for non invasive vibrations measurement

Comunicación presentada en EVACES 2011, 4th International Conference on Experimental Vibration Analysis for Civil Engineering Structures, Varenna (Lecco), Italy, October 3-5, 2011.

GE models and algorithms for condensed phase equilibrium data regression in ternary systems: limitations and proposals

Phase equilibrium data regression is an unavoidable task necessary to obtain the appropriate values for any model to be used in separation equipment design for chemical process simulation and optimization. The accuracy of this process depends on different factors such as the experimental data quality, the selected model and the calculation algorithm. The present paper summarizes the results and conclusions achieved in our research on the capabilities and limitations of the existing GE models and about strategies that can be included in the correlation algorithms to improve the convergence and avoid inconsistencies. The NRTL model has been selected as a representative local composition model. New capabilities of this model, but also several relevant limitations, have been identified and some examples of the application of a modified NRTL equation have been discussed. Furthermore, a regression algorithm has been developed that allows for the advisable simultaneous regression of all the condensed phase equilibrium regions that are present in ternary systems at constant T and P. It includes specific strategies designed to avoid some of the pitfalls frequently found in commercial regression tools for phase equilibrium calculations. Most of the proposed strategies are based on the geometrical interpretation of the lowest common tangent plane equilibrium criterion, which allows an unambiguous comprehension of the behavior of the mixtures. The paper aims to show all the work as a whole in order to reveal the necessary efforts that must be devoted to overcome the difficulties that still exist in the phase equilibrium data regression problem.

Sequential and parallel synchronous alternating iterative methods

The so-called parallel multisplitting nonstationary iterative Model A was introduced by Bru, Elsner, and Neumann [Linear Algebra and its Applications 103:175-192 (1988)] for solving a nonsingular linear system Ax = b using a weak nonnegative multisplitting of the first type. In this paper new results are introduced when A is a monotone matrix using a weak nonnegative multisplitting of the second type and when A is a symmetric positive definite matrix using a P -regular multisplitting. Also, nonstationary alternating iterative methods are studied. Finally, combining Model A and alternating iterative methods, two new models of parallel multisplitting nonstationary iterations are introduced. When matrix A is monotone and the multisplittings are weak nonnegative of the first or of the second type, both models lead to convergent schemes. Also, when matrix A is symmetric positive definite and the multisplittings are P -regular, the schemes are also convergent.

Open-access operating algorithms for commercial videokeratographer and improvement of corneal sampling

We present an algorithm to process images of reflected Placido rings captured by a commercial videokeratoscope. Raw data are obtained with no Cartesian-to-polar-coordinate conversion, thus avoiding interpolation and associated numerical artifacts. The method provides a characteristic equation for the device and is able to process around 6 times more corneal data than the commercial software. Our proposal allows complete control over the whole process from the capture of corneal images until the computation of curvature radii.

GPGPU implementation of growing neural gas: application to 3D scene reconstruction

Self-organising neural models have the ability to provide a good representation of the input space. In particular the Growing Neural Gas (GNG) is a suitable model because of its flexibility, rapid adaptation and excellent quality of representation. However, this type of learning is time-consuming, especially for high-dimensional input data. Since real applications often work under time constraints, it is necessary to adapt the learning process in order to complete it in a predefined time. This paper proposes a Graphics Processing Unit (GPU) parallel implementation of the GNG with Compute Unified Device Architecture (CUDA). In contrast to existing algorithms, the proposed GPU implementation allows the acceleration of the learning process keeping a good quality of representation. Comparative experiments using iterative, parallel and hybrid implementations are carried out to demonstrate the effectiveness of CUDA implementation. The results show that GNG learning with the proposed implementation achieves a speed-up of 6× compared with the single-threaded CPU implementation. GPU implementation has also been applied to a real application with time constraints: acceleration of 3D scene reconstruction for egomotion, in order to validate the proposal.

Parallel relaxed and extrapolated algorithms for computing PageRank

In this paper, parallel Relaxed and Extrapolated algorithms based on the Power method for accelerating the PageRank computation are presented. Different parallel implementations of the Power method and the proposed variants are analyzed using different data distribution strategies. The reported experiments show the behavior and effectiveness of the designed algorithms for realistic test data using either OpenMP, MPI or an hybrid OpenMP/MPI approach to exploit the benefits of shared memory inside the nodes of current SMP supercomputers.