Master Degree / Yüksek Lisans Tezleri

Permanent URI for this collectionhttps://hdl.handle.net/11147/3008

Browse

Filters

2016 - 201952020 - 20234

Settings

Author: search.filters.author.Özuysal, MustafaLanguage: search.filters.language.en

Search Results

Now showing 1 - 9 of 9
  • Master Thesis
    Multi-Frame Super-Resolution Without Priors
    (01. Izmir Institute of Technology, 2023) Gülmez, Veli; Özuysal, Mustafa
    There are mainly two types of super-resolution methods: traditional methods and deep learning methods. While traditional methods define closed-form expressions with assumptions, deep learning methods rely on priors learned from data sets. However, both of them have disadvantages such as being too simple and having strong trust in priors. We focus on how to generate a high-resolution image using low-resolution images without priors by utilizing spatial hash encoding. We propose a grid-based super-resolution model using spatial hash encoding to map coordinate information into higher dimensional space. Our aim is to eliminate long training times and not rely on priors from data sets that are not able to cover all real-world scenarios. Therefore, our proposed model is able to do task- specific super-resolution without priors and eliminate potential hallucination effects caused by wrong priors.
  • Master Thesis
    A Tool for Synthetic Evaluation of Active Calibration Algorithms
    (01. Izmir Institute of Technology, 2022) Dönmez, Buğrahan; Özuysal, Mustafa; Özuysal, Mustafa
    To calibrate a camera, the choice of poses is very important and different angled poses can increase accuracy. Gathering those poses needs expert intuition in order to constrain all parameters accurately. There are various tools to help users calibrate the camera with its guidance. In this study, two successful calibration tools are tested. Both of them guide the user interactively to obtain the best poses. The first method tries to avoid singular poses and captures the poses that reduce the uncertainty of calibration. The second method uses a different approach. It uses the current calibration state to suggest the next pose. In the end, it verifies the parameters with the specified ones by the user. To test these two methods, ground truth data is needed. The ground truth data is obtained with the help of a 3D modeling program. The suggested poses are generated also with the modeling program and knowing the ground truth camera parameters given in the program, the results of the tools are compared.
  • Master Thesis
    Hierarchical Image Classification With Self-Supervised Vision Transformer Features
    (Izmir Institute of Technology, 2022) Karagüler, Caner; Özuysal, Mustafa
    There are lots of works about image classification and most of them are based on convolutional neural networks (CNN). In image classification, some classes are more difficult to distinguish than others because of non-even visual separability. These difficult classes require domain-specific classifiers but traditional convolutional neural networks are trained as flat N-way classifiers. These flat classifiers can not leverage the hierarchical information of the classes well. To solve this issue, researchers proposed new techniques that embeds class-hierarchy into the convolutional neural networks and most of these techniques exceed existing convolutional neural networks' success rates on large-scale datasets like ImageNet. In this work, we questioned if a hierarchical image classification with self- supervised vision transformer features can exceed hierarchical convolutional neural networks. During this work, we used a hierarchical ETHEC dataset and extract attention features with the help of vision transformers. Using these attention features, we implemented 3 different hierarchical classification approaches and compared the results with CNN alternative of our approaches.
  • Master Thesis
    Vehicle Type Classification With Deep Learning
    (Izmir Institute of Technology, 2020) Yaraş, Neriman; Özuysal, Mustafa
    In this thesis, we studied the vehicle type classification problem from several perspectives. We apply a deep learning technique with different parameters such as image size and the number of images in data sets to the classification of an image as one of the nine vehicle types. After choosing the most appropriate one among trained models, we convert the problem into a hierarchical tree classification problem so that it could be analyzed in three different tree hierarchies. Experiments are performed using three computational methods for calculating possibilities for each of the nine classes that correspond to the leaves of the hierarchical trees. These studies result in a conclusion that 0.762812 average accuracy is obtained when traditional arithmetic mean computation applied on the hierarchical tree with level-2 using the Stanford Dataset by 224 image size on ResNet34 architecture.
  • Master Thesis
    Digital font generation using long short-term memory networks
    (Izmir Institute of Technology, 2019) Temizkan, Onur; Özuysal, Mustafa
    Long Short-Term Memory (LSTM) Networks are powerful models to solve sequential problems in machine learning. Apart from their use on sequence classification, LSTMs are also used for sequence prediction. Predictive features of LSTMs have been used extensively to generate handwriting, music and several other types of sequences. Configuration and training of LSTM networks are relatively more arduous than non-sequential models, especially when input data is complex. In this research, the aim is to train LSTM networks and its different variations, use their generative features on a relatively obscure and complex type of sequences in machine learning; digital fonts. Controlled experiments have been performed to find the effects of different model parameters, input encodings or network architectures on learning font based sequences. All in all, in this document; the procedure of creating a dataset from digital fonts are provided, training strategies are demonstrated and the generative results are discussed.
  • Master Thesis
    Detection of Febrile Illness and Its Monitoring Using Iot Technology
    (Izmir Institute of Technology, 2019) Tamur Kaya, Gamze; Tekin, Hüseyin Cumhur; Özuysal, Mustafa
    Measuring and monitoring body temperature has importance for disease diagnosis. It could also help following the course of the disease. Extensive number of body temperature monitoring applications are designed and developed. These applications could be named as healthcare applications aiming to provide convenience to the users. The Internet of Things technology is popular in healthcare applications by offering remote and real-time monitoring. In this project, a telemedicine platform facilitating the diagnosis and monitoring of febrile illnesses is designed. A user-friendly platform is implemented using software components. Message Queuing Telemetry Transport (MQTT), which is a communication protocol, is used for the communication between devices and monitoring system. A broker is used to transmit measured body temperature data from a device to the cloud server. The performance of the broker is evaluated with thousands of generated data packets. It is showed that the platform can handle data requests with a high throughput, i.e, up to 40000 packets/s. The monitoring system is designed as an interactive user interface. For this purpose, Telegram is adapted to the requirements of the platform by using an open-source library of Telegram BOT API. Hence, a user is able to access the measurement results and also control the measurements using instant messaging via Telegram interface. Moreover, this proposed platform could eliminate the use of separate healthcare applications for different diseases by using just a single instant messaging program for all different applications.
  • Master Thesis
    Keypoint Detection and Description on Image Curves
    (Izmir Institute of Technology, 2017) Köksal, Ali; Özuysal, Mustafa
    Image curves are one of the choices for representing interest points which also provide discriminative information about images. Boundary of regions and contour of shapes are real-time instances of image curves. In this thesis, we propose two approaches for keypoint detection and description on image curves. To extract keypoints on image curves, we compute the extrema curvature of region boundaries. This mechanism improves repeatability of keypoints on 3D data. For the description of image curves, shape contours are used. This is similar to approaches that describe the features based on shapes and image gradients. Unlike these approaches, we combine spatial and directional information of tangent directions to extract a feature vector that leads to improved matching and recognition on several standard computer vision tasks such as character and object recognition.
  • Master Thesis
    Keypoint Matching Based on Descriptor Statistics
    (Izmir Institute of Technology, 2016) Uzyıldırım, Furkan Eren; Uzyıldırım, Furkan Eren; Özuysal, Mustafa; Özuysal, Mustafa
    The binary descriptors are the representation of choice for real-time keypoint matching. However, they suffer from reduced matching rates due to their discrete nature. In this thesis, we propose an approach that can augment their performance by searching in the top K near neighbor matches instead of just the single nearest neighbor one. To pick the correct match out of the K near neighbors, we exploit statistics of descriptor bit variations collected for each keypoint individually in an off-line training phase. This is similar in spirit to approaches that learn a patch specific keypoint representation. Unlike these approaches, we limit the use of a keypoint specific score only to rank the list of K near neighbors. Since this list can be efficiently computed with approximate nearest neighbor algorithms, our approach scales well to large descriptor collections.
  • Master Thesis
    Privacy Preservation on Mobile Systems Using Context-Aware Role Based Accss Control
    (Izmir Institute of Technology, 2016) Abdella, Juhar Ahmed; Özuysal, Mustafa; Tomur, Emrah; Özuysal, Mustafa; Tomur, Emrah
    Existing mobile platforms require the user to manually grant and revoke permissions to applications. Once the user grants a given permission to an application, the application can use it without limit unless the user manually revokes the permission. This has become the reason for a lot of privacy problems. One of the solutions suggested by a lot of researchers is Context Aware Access Control (CAAC). However, dealing with policy configurations at permission level becomes very complex as the number of policy rules to configure will become very large. For instance, if there are A applications, P permissions and C contexts, the user may have to deal with A x P x C number of policy configurations. Therefore, we propose a Context-Aware Role-Based Access Control (CA-RBAC) model that can provide dynamic permission granting and revoking while keeping the number of policy rules as small as possible. We demonstrate our model based on Android. In our model, Android applications are assigned roles where roles contain a set of permissions and contexts are associated with permissions. Permissions are activated and deactivated for the containing role based on the associated contexts. Our approach is unique in that our system associates contexts with permissions as opposed to existing similar works which associate contexts with roles. As a proof of concept, we have developed a prototype application called CA-ARBAC (Context-Aware Android Role Based Access Control). We have also performed various tests using our application and the result shows that our model is working as desired.