Traffic and Road Sign Recognition

This thesis presents a system to recognise and classify road and traffic signs for the purpose of developing an inventory of them which could assist the highway engineers’ tasks of updating and maintaining them. It uses images taken by a camera from a moving vehicle. The system is based on three major stages: colour segmentation, recognition, and classification. Four colour segmentation algorithms are developed and tested. They are a shadow and highlight invariant, a dynamic threshold, a modification of de la Escalera’s algorithm and a Fuzzy colour segmentation algorithm. All algorithms are tested using hundreds of images and the shadow-highlight invariant algorithm is eventually chosen as the best performer. This is because it is immune to shadows and highlights. It is also robust as it was tested in different lighting conditions, weather conditions, and times of the day. Approximately 97% successful segmentation rate was achieved using this algorithm.Recognition of traffic signs is carried out using a fuzzy shape recogniser. Based on four shape measures - the rectangularity, triangularity, ellipticity, and octagonality, fuzzy rules were developed to determine the shape of the sign. Among these shape measures octangonality has been introduced in this research. The final decision of the recogniser is based on the combination of both the colour and shape of the sign. The recogniser was tested in a variety of testing conditions giving an overall performance of approximately 88%.Classification was undertaken using a Support Vector Machine (SVM) classifier. The classification is carried out in two stages: rim’s shape classification followed by the classification of interior of the sign. The classifier was trained and tested using binary images in addition to five different types of moments which are Geometric moments, Zernike moments, Legendre moments, Orthogonal Fourier-Mellin Moments, and Binary Haar features. The performance of the SVM was tested using different features, kernels, SVM types, SVM parameters, and moment’s orders. The average classification rate achieved is about 97%. Binary images show the best testing results followed by Legendre moments. Linear kernel gives the best testing results followed by RBF. C-SVM shows very good performance, but ?-SVM gives better results in some case.

CENSUS HISTOGRAMS: A SIMPLE FEATURE EXTRACTION and MATCHING APPROACH FOR FACE RECOGNITION

Most face recognition approaches require a prior training where a given distribution of faces is assumed to further predict the identity of test faces. Such an approach may experience difficulty in identifying faces belonging to distributions different from the one provided during the training. A face recognition technique that performs well regardless of training is, therefore, interesting to consider as a basis of more sophisticated methods. In this work, the Census Transform is applied to describe the faces. Based on a scanning window which extracts local histograms of Census Features, we present a method that directly matches face samples. With this simple technique, 97.2% of the faces in the FERET fa/fb test were correctly recognized. Despite being an easy test set, we have found no other approaches in literature regarding straight comparisons of faces with such a performance. Also, a window for further improvement is presented. Among other techniques, we demonstrate how the use of SVMs over the Census Histogram representation can increase the recognition performance.

Sexual Risk Recognition Deficits: The Role of Prior Victimization and Emotion Dysregulation

Recent theoretical writings suggest that the ineffective regulation of negative emotional states may reduce the ability of women to detect and respond effectively to situational and interpersonal factors that increase risk for sexual assault. However, little empirical research has explored this hypothesis. In the present study, it was hypothesized that prior sexual victimization and negative mood state would each independently predict poor risk recognition and less effective defensive actions in response to an analogue sexual assault vignette. Further, these variables were expected to interact to produce particularly impaired risk responses. Finally, that the in vivo emotion regulation strategy of suppression and corresponding cognitive resource usage (operationalized as memory impairment for the vignette) were hypothesized to mediate these associations. Participants were 668 female undergraduate students who were randomly assigned to receive a negative or neutral film mood induction followed by an audiotaped dating interaction during which they were instructed to indicate when the man had “gone too far” and describe an adaptive response to the situation. Approximately 33.5% of the sample reported a single victimization and 10% reported revictimization. Hypotheses were largely unsupported as sexual victimization history, mood condition, and their interaction did not impact risk recognition or adaptive responding. However, in vivo emotional suppression and cognitive resource usage were shown to predict delayed risk recognition only. Findings suggest that contrary to hypotheses, negative mood (as induced here) may not relate to risk recognition and response impairments. However, it may be important for victimization prevention programs that focus on risk perception to address possible underlying issues with emotional suppression and limited cognitive resources to improve risk perception abilities. Limitations and future directions are discussed.

Deep learning for computer vision: a comparison between convolutional neural networks and hierarchical temporal memories on object recognition tasks

In recent years, Deep Learning techniques have shown to perform well on a large variety of problems both in Computer Vision and Natural Language Processing, reaching and often surpassing the state of the art on many tasks. The rise of deep learning is also revolutionizing the entire field of Machine Learning and Pattern Recognition pushing forward the concepts of automatic feature extraction and unsupervised learning in general. However, despite the strong success both in science and business, deep learning has its own limitations. It is often questioned if such techniques are only some kind of brute-force statistical approaches and if they can only work in the context of High Performance Computing with tons of data. Another important question is whether they are really biologically inspired, as claimed in certain cases, and if they can scale well in terms of "intelligence". The dissertation is focused on trying to answer these key questions in the context of Computer Vision and, in particular, Object Recognition, a task that has been heavily revolutionized by recent advances in the field. Practically speaking, these answers are based on an exhaustive comparison between two, very different, deep learning techniques on the aforementioned task: Convolutional Neural Network (CNN) and Hierarchical Temporal memory (HTM). They stand for two different approaches and points of view within the big hat of deep learning and are the best choices to understand and point out strengths and weaknesses of each of them. CNN is considered one of the most classic and powerful supervised methods used today in machine learning and pattern recognition, especially in object recognition. CNNs are well received and accepted by the scientific community and are already deployed in large corporation like Google and Facebook for solving face recognition and image auto-tagging problems. HTM, on the other hand, is known as a new emerging paradigm and a new meanly-unsupervised method, that is more biologically inspired. It tries to gain more insights from the computational neuroscience community in order to incorporate concepts like time, context and attention during the learning process which are typical of the human brain. In the end, the thesis is supposed to prove that in certain cases, with a lower quantity of data, HTM can outperform CNN.

Recognition of familiar and unfamiliar music in normal aging and Alzheimer's disease

We tested normal young and elderly adults and elderly Alzheimer’s disease (AD) patients on recognition memory for tunes. In Experiment 1, AD patients and age-matched controls received a study list and an old/new recognition test of highly familiar, traditional tunes, followed by a study list and test of novel tunes. The controls performed better than did the AD patients. The controls showed the “mirror effect” of increased hits and reduced false alarms for traditional versus novel tunes, whereas the patients false-alarmed as often to traditional tunes as to novel tunes. Experiment 2 compared young adults and healthy elderly persons using a similar design. Performance was lower in the elderly group, but both younger and older subjects showed the mirror effect. Experiment 3 produced confusion between preexperimental familiarity and intraexperimental familiarity by mixing traditional and novel tunes in the study lists and tests. Here, the subjects in both age groups resembled the patients of Experiment 1 in failing to show the mirror effect. Older subjects again performed more poorly, and they differed qualitatively from younger subjects in setting stricter criteria for more nameable tunes. Distinguishing different sources of global familiarity is a factor in tune recognition, and the data suggest that this type of source monitoring is impaired in AD and involves different strategies in younger and older adults.

Applied Graph Theory in Computer Vision and Pattern Recognition

This book will serve as a foundation for a variety of useful applications of graph theory to computer vision, pattern recognition, and related areas. It covers a representative set of novel graph-theoretic methods for complex computer vision and pattern recognition tasks. The first part of the book presents the application of graph theory to low-level processing of digital images such as a new method for partitioning a given image into a hierarchy of homogeneous areas using graph pyramids, or a study of the relationship between graph theory and digital topology. Part II presents graph-theoretic learning algorithms for high-level computer vision and pattern recognition applications, including a survey of graph based methodologies for pattern recognition and computer vision, a presentation of a series of computationally efficient algorithms for testing graph isomorphism and related graph matching tasks in pattern recognition and a new graph distance measure to be used for solving graph matching problems. Finally, Part III provides detailed descriptions of several applications of graph-based methods to real-world pattern recognition tasks. It includes a critical review of the main graph-based and structural methods for fingerprint classification, a new method to visualize time series of graphs, and potential applications in computer network monitoring and abnormal event detection.

The Role of BDNF, Leptin and Catecholamines in Reward Learning in Bulimia Nervosa

Background: A relationship between bulimia nervosa (BN) and reward-related behavior is supported by several lines of evidence. The dopaminergic dysfunctions in the processing of reward-related stimuli have been shown to be modulated by the neurotrophin brain derived neurotrophic factor (BDNF) and the hormone leptin. Methods: Using a randomized, double-blind, placebo-controlled, crossover design, a reward learning task was applied to study the behavior of 20 female subjects with remitted BN (rBN) and 27 female healthy controls under placebo and catecholamine depletion with alpha-methyl-para-tyrosine (AMPT). The plasma levels of BDNF and leptin were measured twice during the placebo and the AMPT condition, immediately before and 1 h after a standardized breakfast. Results: AMPT-induced differences in plasma BDNF levels were positively correlated with the AMPT-induced differences in reward learning in the whole sample (p = 0.05). Across conditions, plasma BDNF levels were higher in rBN subjects compared to controls (diagnosis effect; p = 0.001). Plasma BDNF and leptin levels were higher in the morning before compared to after a standardized breakfast across groups and conditions (time effect; p < 0.0001). The plasma leptin levels were higher under catecholamine depletion compared to placebo in the whole sample (treatment effect; p = 0.0004). Conclusions: This study reports on preliminary findings that suggest a catecholamine-dependent association of plasma BDNF and reward learning in subjects with rBN and controls. A role of leptin in reward learning is not supported by this study. However, leptin levels were sensitive to a depletion of catecholamine stores in both rBN and controls.

Automated Tools for Layer Recognition (BMPix) and Counting (PEAK)

We present tools for rapid and quantitative detection of sediment lamination. The BMPix tool extracts color and gray-scale curves from images at pixel resolution. The PEAK tool uses the gray-scale curve and performs, for the first time, fully automated counting of laminae based on three methods. The maximum count algorithm counts every bright peak of a couplet of two laminae (annual resolution) in a smoothed curve. The zero-crossing algorithm counts every positive and negative halfway-passage of the curve through a wide moving average, separating the record into bright and dark intervals (seasonal resolution). The same is true for the frequency truncation method, which uses Fourier transformation to decompose the curve into its frequency components before counting positive and negative passages. We applied the new methods successfully to tree rings, to well-dated and already manually counted marine varves from Saanich Inlet, and to marine laminae from the Antarctic continental margin. In combination with AMS14C dating, we found convincing evidence that laminations in Weddell Sea sites represent varves, deposited continuously over several millennia during the last glacial maximum. The new tools offer several advantages over previous methods. The counting procedures are based on a moving average generated from gray-scale curves instead of manual counting. Hence, results are highly objective and rely on reproducible mathematical criteria. Also, the PEAK tool measures the thickness of each year or season. Since all information required is displayed graphically, interactive optimization of the counting algorithms can be achieved quickly and conveniently.

Recognition of Yeast mRNAs as “Nonsense Containing” Leads to Both Inhibition of mRNA Translation and mRNA Degradation: Implications for the Control of mRNA Decapping

A critical step in the degradation of many eukaryotic mRNAs is a decapping reaction that exposes the transcript to 5′ to 3′ exonucleolytic degradation. The dual role of the cap structure as a target of mRNA degradation and as the site of assembly of translation initiation factors has led to the hypothesis that the rate of decapping would be specified by the status of the cap binding complex. This model makes the prediction that signals that promote mRNA decapping should also alter translation. To test this hypothesis, we examined the decapping triggered by premature termination codons to determine whether there is a down-regulation of translation when mRNAs were recognized as “nonsense containing.” We constructed an mRNA containing a premature stop codon in which we could measure the levels of both the mRNA and the polypeptide encoded upstream of the premature stop codon. Using this system, we analyzed the effects of premature stop codons on the levels of protein being produced per mRNA. In addition, by using alterations either in cis or in trans that inactivate different steps in the recognition and degradation of nonsense-containing mRNAs, we demonstrated that the recognition of a nonsense codon led to a decrease in the translational efficiency of the mRNA. These observations argue that the signal from a premature termination codon impinges on the translation machinery and suggest that decapping is a consequence of the change in translational status of the mRNA.

Monoclonal antibodies against oxidized low-density lipoprotein bind to apoptotic cells and inhibit their phagocytosis by elicited macrophages: Evidence that oxidation-specific epitopes mediate macrophage recognition

Apoptosis is recognized as important for normal cellular homeostasis in multicellular organisms. Although there have been great advances in our knowledge of the molecular events regulating apoptosis, much less is known about the receptors on phagocytes responsible for apoptotic cell recognition and phagocytosis or the ligands on apoptotic cells mediating such recognition. The observations that apoptotic cells are under increased oxidative stress and that oxidized low-density lipoprotein (OxLDL) competes with apoptotic cells for macrophage binding suggested the hypothesis that both OxLDL and apoptotic cells share oxidatively modified moieties on their surfaces that serve as ligands for macrophage recognition. To test this hypothesis, we used murine monoclonal autoantibodies that bind to oxidation-specific epitopes on OxLDL. In particular, antibodies EO6 and EO3 recognize oxidized phospholipids, including 1-palmitoyl 2-(5-oxovaleroyl) phosphatidylcholine (POVPC), and antibodies EO12 and EO14 recognize malondialdehyde-lysine, as in malondialdehyde-LDL. Using FACS analysis, we demonstrated that each of these EO antibodies bound to apoptotic cells but not to normal cells, whereas control IgM antibodies did not. Confocal microscopy demonstrated cell-surface expression of the oxidation-specific epitopes on apoptotic cells. Furthermore, each of these antibodies inhibited the phagocytosis of apoptotic cells by elicited peritoneal macrophages, as did OxLDL. In addition, an adduct of POVPC with BSA also effectively prevented phagocytosis. These data demonstrate that apoptotic cells express oxidation-specific epitopes—including oxidized phospholipids—on their cell surface, and that these serve as ligands for recognition and phagocytosis by elicited macrophages.

Intervention of carbohydrate recognition by proteins and nucleic acids.

Carbohydrates in biological systems are often associated with specific recognition and signaling processes leading to important biological functions and diseases. Considerable efforts have been directed toward understanding and mimicking the recognition processes and developing effective agents to control the processes. The pace of discovery research in glycobiology and development of carbohydrate-based therapeutics, however, has been relatively slow due to the lack of appropriate strategies and methods available for carbohydrate-related research. This review summarizes some of the most recent developments in the field, with particular emphasis on work from our laboratories regarding the use of chemoenzymatic strategies to tackle the carbohydrate recognition problem. Highlights include the study of selectin-carbohydrate and aminoglycoside-RNA interactions and development of agents for the intervention of these recognition processes.

Interactions between tRNA identity nucleotides and their recognition sites in glutaminyl-tRNA synthetase determine the cognate amino acid affinity of the enzyme.

Sequence-specific interactions between aminoacyl-tRNA synthetases and their cognate tRNAs both ensure accurate RNA recognition and prevent the binding of noncognate substrates. Here we show for Escherichia coli glutaminyl-tRNA synthetase (GlnRS; EC 6.1.1.18) that the accuracy of tRNA recognition also determines the efficiency of cognate amino acid recognition. Steady-state kinetics revealed that interactions between tRNA identity nucleotides and their recognition sites in the enzyme modulate the amino acid affinity of GlnRS. Perturbation of any of the protein-RNA interactions through mutation of either component led to considerable changes in glutamine affinity with the most marked effects seen at the discriminator base, the 10:25 base pair, and the anticodon. Reexamination of the identity set of tRNA(Gln) in the light of these results indicates that its constituents can be differentiated based upon biochemical function and their contribution to the apparent Gibbs' free energy of tRNA binding. Interactions with the acceptor stem act as strong determinants of tRNA specificity, with the discriminator base positioning the 3' end. The 10:25 base pair and U35 are apparently the major binding sites to GlnRS, with G36 contributing both to binding and recognition. Furthermore, we show that E. coli tryptophanyl-tRNA synthetase also displays tRNA-dependent changes in tryptophan affinity when charging a noncognate tRNA. The ability of tRNA to optimize amino acid recognition reveals a novel mechanism for maintaining translational fidelity and also provides a strong basis for the coevolution of tRNAs and their cognate synthetases.