This thesis investigates the performance of Mask R-CNN when utilizing transfer learning on a small dataset. The aim was to instance segment eyeglass lenses as accurately as possible from self-portrait images. Five different models were trained, where the key difference was the types of eyeglasses the models were trained on. The eyeglasses were grouped into three types, fully rimmed, semi-rimless, and rimless glasses. 1550 images were used for training, validation, and testing. The model's performances were evaluated using TensorBoard training data and mean Intersection over Union scores (mIoU). No major differences in performance were found in four of the models, which grouped all three types of glasses into one class. Their mIoU scores range from 0.913 to 0.94 whereas the model with one class for each group of glasses, performed worse, with a mIoU of 0.85. The thesis revealed that one can achieve great instance segmentation results using a limited dataset when taking advantage of transfer learning. / Denna uppsats undersöker prestandan för Mask R-CNN vid användning av överföringsinlärning på en liten datamängd. Syftet med arbetet var att segmentera glasögonlinser så exakt som möjligt från självporträttbilder. Fem olika modeller tränades, där den viktigaste skillnaden var de typer av glasögon som modellerna tränades på. Glasögonen delades in i 3 typer, helbåge, halvbåge och båglösa. Totalt samlades 1550 träningsbilder in, dessa annoterades och användes för att träna modellerna.  Modellens prestanda utvärderades med TensorBoard träningsdata samt genomsnittlig Intersection over Union (IoU). Inga större skillnader i prestanda hittades mellan modellerna som endast tränades på en klass av glasögon. Deras genomsnittliga IoU varierar mellan 0,913 och 0,94. Modellen där varje glasögonkategori representerades som en unik klass, presterade sämre med en genomsnittlig IoU på 0,85. Resultatet av uppsatsen påvisar att goda instanssegmenteringsresultat går att uppnå med hjälp av en begränsad datamängd om överföringsinlärning används.

No description available.

En los últimos años, con el surgimiento de sistemas multimedia se ha vuelto popular la incorporación de cámaras con proce-samiento en el mismo chip, en productos de consumo como cámaras de video, cámaras fotográficas, teléfonos celulares, reproductores multimedia, etc. En esta tesis se presenta el analisis de una arquitecturas que permite crear una "cámara inteligente" que incorpora la capacidad de procesar las imá-genes que adquiere utilizando un procesamiento paralelo distribuido sobre el plano focal. El funcionamiento se basa en una estructura del tipo CNN simplicial, donde cada celda opera en función de su información y la de celdas vecinas. Cada celda implementa una ecuación discreta de evolución de estado, basada en una función lineal por tramos multidimen-sional. Las celdas se programan a través de una única memo-ria que se dispone en la periferia del integrado, y el cálculo se realiza con señales codificadas en tiempo, lo cual permite una realización muy eficiente desde el punto de vista del área ocupada por cada celda. Se presentan dos circuitos integrados diseñados bajo estos principios. Se han fabricado dos circui-tos integrados, el primero en una tecnología CMOS estándar de 90nm que contiene un arreglo de 64 x 64 celdas. El segundo se diseñó en una tecnología 3D de dos pisos de 0; 13pm y contiene un arreglo de 48 x 32 celdas. / In recent years, with the emergence of Multimedia systems cameras with onchip processing has become popular, in consumer products like video cameras, cell phones, media players, etc. This thesis presents the analysis of an archi-tecture of a "smart" camera that has the ability of acquire and to process images using a parallel processing. This chip works based on a simplicial cnn structure, where each cell operates according to the neighborhood information. Each cell implements a discrete state equation, based on a multidimen-sional piecewise linear function. The cells are programmed with memory on the periphery of the integrated, and the calculation is performed with time coded signals, which allows very eficient realization in terms of area. Two integrated circuits are presented here, designed under these principles. The first is 64 times 64 array fabricated on a 90nm CMOS technology. The second was designed in a 3D 0;13 mum technology and contains an array of 48 times 32cells.

Bearings are critical in today’s mechanisms, and their reliability is continuously improving. Yet, working under high loads for long periods, bearings will degrade and eventually fail. An unpredicted bearing failure can lead to total and catastrophic failures of machines and may even lead to human injuries that result in substantial economic losses and reductions in production. Determining a bearing’s remaining useful life (RUL) has become an important topic in many industrial fields. Vibration signals are the most used representation for understanding a bearing’s health status. Using different algorithms, time-domain vibration signals can be transformed into time-frequency domain signals that help indicate a bearing’s status. For instance, this thesis investigates spectrograms and scalograms to visually represent a bearing’s health condition using a short-time Fourier transform (STFT) and a continuous wavelet transform (CWT). Both representations are plotted as a function of time and frequency and can detect the bearing’s working condition. However, spectrograms are advantageous in revealing frequency changes along the time axis, while scalograms facilitate the detection of abrupt changes. Combined with a convolutional neural network (CNN), these plots can be used to interpret bearing RUL. The strength of CNNs lie in their ability to identify and detect features in images, including such tasks as image classification, using share-weight architectures, convolutional layers, and kernels. This thesis explores CNNs combined with spectrograms and scalograms using the PRONOSTIA dataset to perform bearing RUL predictions and explore relationships between prognosis and diagnosis for bearing faults analysis.

Due to the rapid progress and advancements in deep learning and neural networks, manyapproaches and state-of-the-art researches have been conducted in these fields which cause developing various learning-based attacks leading to vulnerability of websites and portals. This kind of attacks decrease the security of the websites which results in releasing the sensitive and important personal information. These days, preserving the security of the websites is one of the most challenging tasks. CAPTCHA (Completely Automated Public Turing Test to Tell Computers and Humans Apart) is kind of test which are developed by designers and are available in various websites to distinguish and differentiate humans from robots in order to protect the websites from possible attacks. In this dissertation, we proposed a CNN based approach to attack and break text-based CAPTCHAs. The proposed method has been compared with several state-of-the-art approaches in terms of recognition accuracy (RA). Based on the results, the developed method can break and recognize CAPTCHAs at high accuracy. Additionally, we wanted to check how to make these CAPTCHAs hard to be broken, so we employed five types of distortions in these CAPTCHAs. The recognition accuracy in presence of these noises has been calculated. The results indicate that adversarial noise can make CAPTCHAs much difficult to be broken. The results have been compared with some state-of-the-art approaches. This analysis can be helpful for CAPTCHA developers to consider these noises in their developed CAPTCHAs. This dissertation also presents a hybrid model based on CNN-SVM to solve text-based CAPTCHAs. The developed method contains four main steps, namely: segmentation, feature extraction, feature selection, and recognition. For segmentation, we suggested using histogram and k-means clustering. For feature extraction, we developed a new CNN structure. The extracted features are passed through the mRMR algorithm to select the most efficient features. These selected features are fed into SVM for further classification and recognition. The results have been compared with several state-of-the-art methods to show the superiority of the developed approach. In general, this dissertation presented deep learning-based methods to solve text-based CAPTCHAs. The efficiency and effectiveness of the developed methods have been compared with various state-of-the-art methods. The developed techniques can break CAPTCHAs at high accuracy and also in a short time. We utilized Peak Signal to Noise Ratio (PSNR), ROC, accuracy, sensitivity, specificity, and precision to evaluate and measure the performance analysis of different methods. The results indicate the superiority of the developed methods.

No description available.

The capacity to estimate the number of objects (numerosity) in the environment is ontogenetically precocious and phylogenetically ancient. In animals, this ability holds significant adaptive advantages, directly influencing survival and reproductive success. In humans, it may serve an additional purpose by providing a start-up kit for the acquisition of symbolic numbers, thus making it a potential focus for mathematics education and intervention strategies. Behavioral, neurophysiological, and neuroimaging findings suggest that numerosity information is directly extracted from the environment. However, numerosity is inherently linked with other visual characteristics of sets (such as larger sets often occupy more space or are more densely spaced), making it challenging to determine the extent to which the observed response to numerosity is distinct from the response to other visual attributes. In my PhD research I provide experimental evidence through neuroimaging and computational modeling techniques elucidating where, when, and how numerical information is encoded in the human brain. This work therefore provides a threefold contribution. First, I show that numerosity is represented over and above nonnumeric visual features in a widespread network of areas starting from early visual areas and further amplified in associative areas along the dorsal but also notably the ventral stream, and that the neural representational geometries of regions across the two steams are substantially identical. Second, I showed that numerosity is represented at an early stage and seemingly in parallel across of a set of regions including early visual, parietal, and temporal, preceding the emergence of non-numeric features that could indirectly contribute to numerosity computation. Finally, by comparing the fMRI data with a convolutional neural network (CNN) to explore similarities and differences between the model and human brain data, I discovered that although the CNN can perform approximate numerosity comparisons and the structure of their representation in their hidden layers captures well numerosity representation in early visual areas of humans, it falls short of fully simulating the way in which associative brain regions represent numerosity. Taken together, the findings of this thesis provide experimental evidence supporting the notion that number is a primary visual feature, encoded independent from other visual features quickly and widely across the human brain. Furthermore, they emphasize the need for additional investigation to unravel the computational mechanisms underlying numerosity in the human brain.

Sistemas de recuperação de imagens por conteúdo (do inglês Content-based ImageRetrieval - CBIR) têm sido cada vez mais utilizados em diversas aplicações de tratamento e análise de imagens, devido a dois fatores: CBIR é um procedimento que pode ser feito automaticamente, permitindo tratar o grande volume de imagens adquiridos em hospitais, e também é a base para o processamento de consultas por similaridade. No contexto médico tais sistemas auxiliam em diversas tarefas, desde treinamento de profissionais até em sistemas de auxílio a diagnóstico (do inglês Computer-Aided Diagnosis - CAD). Um sistema computacional capaz de comparar e classificar imagens obtidas em exames de pacientes utilizando uma base prévia de conhecimento poderia agilizar o atendimento da população e fornecer aos especialistas informações relevantes de forma rápida e simples. Neste trabalho, o foco foi na análise de imagens de úlceras venosas. Foram desenvolvidas duas técnicas para classificação dessas imagens. A primeira, denominada Counting-Labels Similarity Measure (CL-Measure) possuia vantagem de lidar com imagens segmentadas de forma automática, por superpixels, e ser versátil o suficiente para permitir adaptação para outros domínios. A ideia principal do CL-Measure consiste na criação de sub-imagens baseadas em uma classificação prévia, calcular a distância entre elas e agregar as distâncias parciais obtidas a partir de uma função apropriada. A segunda técnica, denominada Quality of Tissues from Dermatological Ulcers(QTDU), faz uso de redes convolucionais (CNNs) para rotulação dos superpixels com a vantagem de compor todo o processo de identificação de características e classificação, dispensando a necessidade de identificar qual o extrator de características mais adequado para o contexto em questão. Experimentos realizados sobre a base de imagens analisada, utilizando 179572 super pixels divididos em 4 classes, indicam que a QTDU é a abordagem mais eficaz até o momento para o contexto de classificação de imagens dermatológicas, com médias de AUC=0,986, sensitividade = 0,97,e especificidade=0,974 superando as abordagens anteriores baseadas em aprendizado de máquina em 11;7% e 8;2% considerando o coeficiente KAPPAeF-Measure, respectivamente. / Content-based Image Retrieval (CBIR) systems have been increasingly used in many image processing and analysis applications because of two factors: CBIR is a procedure that can be done automatically, allowing to handle the large volume of images acquired in hospitals, and it is also the basis for processing similarity queries. In the medical context, such systems assist in various tasks, from training of professionals to develop Computer-Aided Diagnosis CAD systems. A computer system capable of comparing and classifying images obtained from patient exams using a prior knowledge base could expedite the care of the population and provide specialists with relevant information quickly. In this study, the focus was on the analysis of images of venous ulcers. Two techniques were developed to classify these images. The first, called Counting-Labels Similarity Measure (CL-Measure) has the advantage of dealing with automatically segmented images by superpixels, and is versatile enough to allow adaptation to other domains. The main idea of CL-Measure is to create sub-images based on a previous classification, calculate the distance between them and add the partial distances obtained from an appropriate function. The second technique, called Quality of Tissues from Dermatological Ulcers (QTDU), makes use of convolutional networks (CNNs) for superpixels labeling, with the advantage of encompassing the whole process of identification of features and classification, without the need of identifying which extractor would be the best for the context in question. Experiments carried out on the image database using 179,572 superpixels divided into 4 classes, indicate that the QTDU is the most effective approach to date for the context of classification of dermatological ulcer images, with averages of AUC = 0.986, sensitivity = 0.97 , and specificity = 0.974, surpassing previous approaches based on machine learning in 11.7% and 8.2% considering the KAPPA and F-Measure coefficients, respectively.

Diploma thesis is concerned with the topic of relations between modern media and foreign policy. In a general context of this relation, we focused on the theory of CNN effect, describing it causes and particular manifestations mutual media and policy interactions. We try to analyze CNN effect on single case of armed conflict in Libya and its media coverage by influential American daily papers. By combination of existing scientific methods, we created adapted research method for our case, which is based on quantitative content analyses of all published articles with main topics - armed conflict in Libya - in the period of 33 days till the adoption of Security Council resolution. Main variables of our research was the date of publishing, articles range, section in newspaper and especially "value direction" of article in favor or not in favor of armed intervention to Libya. Analyses of media coverage of this topic were then confronted with factual analyses of topic in same time period. Results of our research confirmed potential relation between media coverage and US state department decision to intervene in Libya conflict. For absolute causality confirmation between these two variables would be necessary much more extensive research. Additional output of this research was confirmation of partial...

Zugl.: Dresden, Techn. Univ., Magisterarbeit