Biological vision and computational models of vision can be split into three independent components (image description, decision process, and image set). The thesis presented here aimed to investigate the influence of each of these core components on computational model’s similarity to human behaviour. Chapter 3 investigated the similarity of different computational image descriptors to their biological counterparts, using an image matching task. The results showed that several of the computational models could explain a significant amount of the variance in human performance on individual images. The deep supervised convolutional neural net explained the most variance, followed by GIST, HMAX and then PHOW. Chapter 4 investigated which computational decision process best explained observers’ behaviour on an image categorization task. The results showed that Decision Bound theory produced behaviour the closest to that of observers. This was followed by Exemplar theory and Prototype theory. Chapter 5 examined whether the naturally differing image set between computational models and observers could partially account for the difference in their behaviour. The results showed that, indeed, the naturally differing image set between computational models and observers was affecting the similarity of their behaviour. This gap did not alter which image descriptor best fit observers’ behaviour and could be reduced by training observers on the image set the computational models were using. Chapter 6 investigated, using computational models of vision, the impact of the neighbouring (masking) images on the target images in a RSVP task. This was done by combining the neighbouring images with the target image for the computational models’ simulation for each trial. The results showed that models behaviour became closer to that of the human observers when the neighbouring mask images were included in the computational simulations, as would be expected given an integration period for neural mechanisms. This thesis has shown that computational models can show quite similar behaviours to human observers, even at the level of how they perform with individual images. While this shows the potential utility in computational models as a tool to study visual processing, It has also shown the need to take into account many aspects of the overall model of the visual process and task; not only the image description, but the task requirements, the decision processes, the images being used as stimuli and even the sequence in which they are presented.

Handling large displacement optical flow is a remarkably arduous task. For instance, standard coarse-to-fine techniques often struggle to adequately deal with moving objects whose motion exceeds their size. Here we propose a learning approach to the estimation of large displacement between two non-consecutive images in a sequence on the basis of a learning set of optical flows estimated a priori between different consecutive images in the same sequence. Our method refines an initial estimate of the flow field by replacing each displacement vector by a linear combination of displacement vectors at the center of similar patches taken from a code-book built from the learning set. The key idea is to use the accurate flows estimated a priori between consecutive images to help improve the potentially less accurate flows estimated online between images further apart. Experimental results suggest the ability of a purely data-driven learning approach to handle fine scale structures with large displacements.

Ultrawideband (UWB) technology has received considerable attention in recent years as it is regarded to be able to revolutionise a wide range of applications. UWB imaging for breast cancer detection is particularly promising due to its appealing capabilities and advantages over existing techniques, which can serve as an early-stage screening tool, thereby saving millions of lives. Although a lot of progress has been made, several challenges still need to be overcome before it can be applied in practice. These challenges include accurate signal propagation modelling and breast phantom construction, artefact resistant imaging algorithms in realistic breast models, and low-complexity implementations. Under this context, novel solutions are proposed in this thesis to address these key bottlenecks. The thesis first proposes a versatile electromagnetic computational engine (VECE) for simulating the interaction between UWB signals and breast tissues. VECE provides the first implementation of its kind combining auxiliary differential equations (ADE) and convolutional perfectly matched layer (CPML) for describing Debye dispersive medium, and truncating computational domain, respectively. High accuracy and improved computational and memory storage efficiency are offered by VECE, which are validated via extensive analysis and simulations. VECE integrates the state-of-the-art realistic breast phantoms, enabling the modelling of signal propagation and evaluation of imaging algorithms. To mitigate the severe interference of artefacts in UWB breast cancer imaging, a robust and artefact resistant (RAR) algorithm based on neighbourhood pairwise correlation is proposed. RAR is fully investigated and evaluated in a variety of scenarios, and compared with four well-known algorithms. It has been shown to achieve improved tumour detection and robust artefact resistance over its counterparts in most cases, while maintaining high computational efficiency. Simulated tumours in both homogeneous and heterogeneous breast phantoms with mild to moderate densities, combined with an entropy-based artefact removal algorithm, are successfully identified and localised. To further improve the performance of algorithms, diverse and dynamic correlation weighting factors are investigated. Two new algorithms, local coherence exploration (LCE) and dynamic neighbourhood pairwise correlation (DNPC), are presented, which offer improved clutter suppression and image resolution. Moreover, a multiple spatial diversity (MSD) algorithm, which explores and exploits the richness of signals among different transmitter and receiver pairs, is proposed. It is shown to achieve enhanced tumour detection even in severely dense breasts. Finally, two accelerated image reconstruction mechanisms referred to as redundancy elimination (RE) and annulus predication (AP) are proposed. RE removes a huge number of repetitive operations, whereas AP employs a novel annulus prediction to calculate millions of time delays in a highly efficient batch mode. Their efficacy is demonstrated by extensive analysis and simulations. Compared with the non-accelerated method, RE increases the computation speed by two-fold without any performance loss, whereas AP can be 45 times faster with negligible performance degradation.

<p>This thesis describes the research I have done, and partly will do, during my time as a PhD student in the laboratory of Applied Optics at Linköping University. Due to circumstances beyond the scope of this book, this incorporates three quite different projects. The first two, involving gas sensing and measuring on paper with ellipsometry, have been discontinued, whereas the third one, measuring fluorescence with a computer screen and web camera, is in full progress and will be until I complete my studies.</p><p>Thus the purpose of this work also has several aspects. Partly, it describes performed research and its results, as well as theoretical background. On the other hand, it provides practical and theoretical background necessary for future work. While the three projects are truly quite different, each of them has certain things in common with each of the other. This is certainly also true for the necessary theory. Two of them involve spectroscopic ellipsometry, for example, while another pair needs knowledge of color theory, etc. This makes it impossible to separate the projects, despite of their differences. Hopefully, these links between the different projects, connecting the different chapters, will make this work whole and consistent in its own way.</p> / Report code: LiU-TEK-LIC-200 4-19. On the day of the public defence the status of article I was: In press and the status of article III was: Manuscript and has a new title. The old title was Computer screen photo-assisted spectroscopic fluorimetry.

This thesis studies the relationship between the pairs of resonator output coupling and intra-cavity absorber initial transmission, and the FWHM (full width at half maximum) pulse duration of a continuously pumped passively Q-switched solidstate laser, when the output energy is pre-determined. Depending on the magnitude of the pumping power, three different rate equation models are used to evaluate the required output coupler reflectivity and absorber initial-transmission pair for the corresponding FWHM pulse duration. The energy transfer kinetics of the passively Q-switched laser decides the required pumping power; and the pair of output coupler reflectivity absorber transmission pair, determine the build-up time of Q-switching and the repetition rate of the laser system. Hence, the forms of the models are controlled by two conditions: 1) the build-up time of Q-switching; and 2) the recovery time of the absorber. When the build-up time of Q-switching is relatively short, but the recovery time of the absorber is long, Model I is based on the simplified laser rate equations. It is used to evaluate the output coupler reflectivity and absorber initialtransmission pair, which satisfies the pre-determined output energy and FWHM pulse duration. Model II is set up to study the case when both the build-up time of Q-switching and the recovery time of the absorber are long. In Model II, the laser rate equations are solved using the Runge-Kutta method. Model III simulates the case when the recovery time of the absorber is short. To validate the models, the simulation results of practical passively Q-switched laser systems are compared with experimental results reported in the literature. The agreement of the simulation results with reported experimental results demonstrates the importance of the boundary conditions for the different cases, and verifies the soundness of the models. Generalizing the simulation results, obtained from different passively Q-switched laser systems with different pumping power and different pre-determined output energy, yields general conclusions which permit a designer to select the correct parameters for a desired laser performance.

Metal nanoparticles have unique optical, electrical, catalytic and mechanical properties, which lead them to various applications in nanotechnology. In particular, noble metal nanowires are attracting growing attention due to their potential applications such as in opto-electronic devices and transparent conductive contacts (TCCs). There are two general approaches to synthesize nanowires: template-assisted and solution phase methods. However, these synthesis approaches have various disadvantages. For example, removal of the template to ensure the purity of the synthesized nanowires is the major problem. In solution methods like the widely used &ldquo / seed mediated growth method&rdquo / , nanowires are synthesized in low yield with the significant amount of by-products and requirement of purification is a major problem for further applications. Among all solution based methods, hydrothermal process is a very promising way of preparing gold (Au) nanowires in high yield and structural purity. In this thesis, hydrothermal process was modified to synthesize high aspect ratio Au nanowires with high morphological purity. Parametric study was performed to examine the effect of surfactant concentration, reaction time and temperature on the quality of products. The optimum conditions were determined for two different surfactant molecules (hexamethylenetetramine (HMTA) and ethylenediaminetetraacetic acid (Na2-EDTA)). Characterization of the products was done by detailed analysis via scanning electron microscopy (SEM), transmission electron microscopy (TEM), energy-dispersive spectroscopy (EDS), X-Ray diffraction (XRD) and X-Ray photoelectron spectroscopy (XPS). The analyses demonstrated that the Au nanowires synthesized at optimum conditions have high aspect ratio (diameters 50-110 nm range and lengths in micrometer range) and high structural purity.

Computer technology has rapidly improved over the last few years, with more powerful machines becoming ever smaller and cheaper. The latest growth area is in portable personal computers, providing powerful facilities to the mobile business person. Alongside this development has been the vast improvement to the human computer interface, allowing noncomputer- literate users access to computing facilities. These two aspects are now being combined into a portable computer that can be operated with a stylus, without the need for a keyboard. Handwriting is the obvious method for entering data and cursive script recognition research aims to comprehend unconstrained, natural handwriting. The ORCHiD system described in this thesis recognises connected handwriting collected on-line, in real time, via a digitising pad. After preprocessing, to remove any hardware-related errors, and normalising, the script is segmented and features of each segment measured. A new segmentation method has been developed which appears to be very consistent across a large number of handwriting styles. A statistical template matching algorithm is used to identify the segments. The system allows ambiguous matching, since cursive script is an ambiguous communications medium when taken out of context, and a probability for each match is calculated. These probabilities can be combined across the word to produce a ranked list of possible interpretations of the script word. A fast dictionary lookup routine has been developed enabling the sometimes very large list of possible words to be verified. The ORCHiD system can be trained, if desired, to a particular user. The training routine, however, is automatic since the untrained recognition system is used as the basis for the trained system. There is therefore very little start-up time before the system can be used. A decision-directed training approach is used. Recognition rates for the system vary depending on the consistency of the writing. On average, the untrained system achieved 75% recognition. After some training, average recognition rates of 91% were achieved, with up to 96% observed after further training.

L’injection tout optique des porteurs de charges libres dans un matériau semi-conducteur peut changer la fréquence de résonance d’une cavité optique pendant quelques picosecondes et permet une modification ultra-rapide de l’interaction lumière-matière. Dans cette thèse, nous étudions la commutation de différents types de cavités basés sur les matériaux GaAs/AlAs et explorons des applications possibles.Quand la longueur d’onde de résonance est changée sur une échelle de temps inférieure à son temps de stockage, lumière stockée subit un décalage vers les hautes fréquences. Dans ce travail, nous étudions expérimentalement cet effet pour des microcavités planaires à haut facteur de qualité, étant capables de stocker la lumière pendant plusieurs dizaines de picosecondes et observons un décalage important (environ 17 largeurs de raie) de la lumière stockée. Conformément à nos simulations numériques, nous mettons en évidence un comportement adiabatique et une efficacité proche de 100% pour ce procédé de conversion.Étant au sein d’une cavité, des boîtes quantiques (BQ) peuvent servir comme source de lumière interne pour sonder les modes de la cavité et la dynamique de la commutation. Nous utilisons cette approche pour étudier deux types de cavités différents.D’un côté, nous injectons une distribution inhomogène de porteurs de charge libres dans des micropiliers, dont l’intérêt pour des expériences d’optique quantique est bien reconnu. A cause des recouvrements différents entre les distributions des porteurs libres et des intensités des champs, nous observons des comportements de la commutation des modes radicalement différents. Ce comportement est compris quantitativement sur la base des simulations prenant en compte la diffusion et la recombinaison des paires électron-trou.D’un autre côté, nous explorons les propriétés d’un nouveau type de microcavité, des résonateurs en forme d’anneau ovoïde. Nous présentons une caractérisation de leurs propriétés optiques et des expériences de commutation. Ces objets présentent des perspectives prometteuses pour la fabrication des microlasers et pour des expériences d’optique, telles que le contrôle dynamique de l’effet Purcell. / The all-optical injection of free charge carriers into a semiconductor material can change the resonance frequency of an optical microcavity within few picoseconds and allows an ultrafast modification of light-matter interaction. In this PhD thesis, we study the switching of different types of cavities based on GaAs/AlAs materials and explore possible applications.When the resonance wavelength of a cavity is shifted on a timescale shorter than its storage time, the frequency of the stored light is up-converted. In this work, we study this effect experimentally for high Q planar microcavities, able to store light during several tens of picoseconds. Upon ultrafast switching, we observe a large frequency shift (around 17 mode linewidths) of stored light.In agreement with numerical simulations, we evidence an adiabatic behavior and an efficiency close to 100% for this conversion process.When embedded in a cavity, quantum dots can serve as an internal light source for probing cavity modes and their switching dynamics. We use this approach to study two different kinds of microcavities.On one hand, we inject an inhomogeneous distribution of free charge carriers into micropillars, whose interest for quantum optics experiments is well recognized. We observe drastically different switching behaviors for their cavity modes, due to the different overlaps between free carriers and field intensity distributions. This behavior is understood in a quantitative way on the basis of simulations taking into account the diffusion and recombination of electron-hole pairs.On the other hand, we explore the properties of a novel type of microcavity, ovoid ring resonators. We present a characterization of their optical properties, as well as switching experiments. These objects offer appealing perspectives for the fabrication of microlasers, and for quantum optics experiments such as controlling the Purcell effect in real time.

Facial Landmark Tracking (Face Tracking) is a key step for many Face Analysis systems, such as Face Recognition, Facial Expression Recognition, or Age and Gender Recognition, among others. The goal of Facial Landmark Tracking is to locate a sparse set of points defining a facial shape in a video sequence. These typically include the mouth, the eyes, the contour, or the nose tip. The state of the art method for Face Tracking builds on Cascaded Regression, in which a set of linear regressors are used in a cascaded fashion, each receiving as input the output of the previous one, subsequently reducing the error with respect to the target locations. Despite its impressive results, Cascaded Regression suffers from several drawbacks, which are basically caused by the theoretical and practical implications of using Linear Regression. Under the context of Face Alignment, Linear Regression is used to predict shape displacements from image features through a linear mapping. This linear mapping is learnt through the typical least-squares problem, in which a set of random perturbations is given. This means that, each time a new regressor is to be trained, Cascaded Regression needs to generate perturbations and apply the sampling again. Moreover, existing solutions are not capable of incorporating incremental learning in real time. It is well-known that person-specific models perform better than generic ones, and thus the possibility of personalising generic models whilst tracking is ongoing is a desired property, yet to be addressed. This thesis proposes Continuous Regression, a Functional Regression solution to the least-squares problem, resulting in the first real-time incremental face tracker. Briefly speaking, Continuous Regression approximates the samples by an estimation based on a first-order Taylor expansion yielding a closed-form solution for the infinite set of shape displacements. This way, it is possible to model the space of shape displacements as a continuum, without the need of using complex bases. Further, this thesis introduces a novel measure that allows Continuous Regression to be extended to spaces of correlated variables. This novel solution is incorporated into the Cascaded Regression framework, and its computational benefits for training under different configurations are shown. Then, it presents an approach for incremental learning within Cascaded Regression, and shows its complexity allows for real-time implementation. To the best of my knowledge, this is the first incremental face tracker that is shown to operate in real-time. The tracker is tested in an extensive benchmark, attaining state of the art results, thanks to the incremental learning capabilities.

Habitat classification, the process of mapping a site with its habitats, is a crucial activity for monitoring environmental biodiversity. Phase 1 classification, a 10-class four-tier hierarchical scheme, is the most widely used scheme in the UK. Currently, no automatic approaches have been developed and its classification is carried out exclusively by ecologists. This manual approach using surveyors is laborious, expensive and subjective. To this date, no automatic approach has been developed. This thesis presents the first automatic system for Phase 1 classification. Our main contribution is an Automatic Image Annotation (AIA) framework for the automatic classification of Phase 1 habitats. This framework combines five elements to annotate unseen photographs: ground-taken geo-referenced photography, low-level visual features, medium-level semantic information, random projections forests and location-based weighted predictions. Our second contribution are two fully-annotated ground-taken photograph datasets, the first publicly available databases specifically designed for the development of multimedia analysis techniques for ecological applications. Habitat 1K has over 1,000 photographs and 4,000 annotated habitats and Habitat 3K has over 3,000 images and 11,000 annotated habitats. This is the first time ground-taken photographs have been used with such ecological purposes. Our third contribution is a novel Random Forest-based classifier: Random Projection Forests (RPF). RPFs use Random Projections as a dimensionality reduction mechanism in their split nodes. This new design makes their training and testing phase more efficient than those of the traditional implementation of Random Forests. Our fourth contribution arises from the limitations that low-level features have when classifying similarly visual classes. Low-level features have been proven to be inadequate for discriminating high-level semantic concepts, such as habitat classes. Currently, only humans posses such high-level knowledge. In order to obtain this knowledge, we create a new type of feature, called medium-level features, which use a Human-In-The-Loop approach to extract crucial semantic information. Our final contribution is a location-based voting system for RPFs. We benefit from the geographical properties of habitats to weight the predictions from the RPFs according to the geographical distance between unseen test photographs and photographs in the training set. Results will show that ground-taken photographs are a promising source of information that can be successfully applied to Phase 1 classification. Experiments will demonstrate that our AIA approach outperforms traditional Random Forests in terms of recall and precision. Moreover, both our modifications, the inclusion of medium-level knowledge and a location-based voting system, greatly improve the recall and precision of even the most complex habitats. This makes our complete image-annotation system, to the best of our knowledge, the most accurate automatic alternative to manual habitat classification for the complete categorization of Phase 1 habitats.