|
|
Impact Of Loss Function On Multi-Frame Super-Resolution
Mezina, Anzhelika
Nowadays, one of the most popular topics in image processing is super-resolution. Thisproblem is getting more actual even in security, since monitoring cameras are everywhere and inthe case of an incident, it is necessary to recognize a person from records. A lot of approaches exist,which are able to reconstruct image, and the most of them are based on deep learning. The main focusof this work is to analyze, which loss function for neural networks is more effective for real-worldimage reconstruction. For this experiment chosen architecture and dataset are used for multi-framesuper-resolution for _x0002_8 scaling.
|
|
|
Online database for secure data collection
Kopec, Peter ; Mezina, Anzhelika (oponent) ; Mikulec, Marek (vedoucí práce)
This bachelor thesis deals with the design and implementation of a secure online database for data collection, which is accessible from the Internet. A database that is accessible from the Internet and contains personal data or other valuable data must be well secured, because we do not want this data to be misused by an unauthorized person. To begin with, we select the appropriate applications for our system and analyze their functionality. The applications are selected based on the features they provide, the overall complexity and support of their online community. Part of the work is devoted to the analysis of data leaks from medical facilities in 2019 and 2020 and a few other leaks from other industries. Thanks to this analysis, we know the reasons for the data leakage and we are able to focus more on these weaknesses and point out the problems. The next part of the work is devoted to the design and implementation of a practical solution using applications that we selected at the beginning. In our case it is a MYSQL database, FLASK backend with Gunicorn WSGI and NGINX web server. Finally, we analyze the security of this solution using the most common vulnerabilities according to OWASP and the NMAP network scanner.
|
|
|
Increasing quality of facial images using sequence of images
Svorad, Adam ; Mezina, Anzhelika (oponent) ; Burget, Radim (vedoucí práce)
Master’s thesis delves into the field of face super-resolution. It aims to review novel approaches to single-frame image sharpening and image editing in the theoretical part of the work. Practical part will focus on approaches to image reconstruction from a sequence of damaged images. Multiple multi-frame neural network models will be implemented and evaluated. As alternative option, a suite of image editing tools will be presented as well. These tools will utilize most modern image editing techniques to merge visual features of faces from multiple input images into a single output image. At the end of the thesis, all methods will be compared to each other.
|
|
|
Face superresolution from image sequence
Mezina, Anzhelika ; Rajnoha, Martin (oponent) ; Burget, Radim (vedoucí práce)
This work is focused on application of deep learning in increasing resolution of images containing face. This can be applied in different fields, including security. For example, in case of incident, the police needs to identify a culprit from the records of security camera. The aim of this work is to propose neural network models, which would work with sequence of frames, and to compare these models with existing methods for a single image super-resolution. For this purpose, a new dataset with sequences of the images with faces is created. The methods for the single super-resolution are trained on the new dataset. The new architectures for multiframe super-resolution are proposed. They are based on U-Net model. This model is successful for segmentation tasks, but it can be also applied for super-resolution tasks. To improve this architecture, the residual blocks and its modification are used. To avoid blurring effect and recover more details, the perceptual loss function is applied. In the first part of this work, the description of neural networks and overview of the architectures, which can be applied in super-resolution, is provided. The second part contains the methods for super-resolution of a single frame, multiframe, video. In the next section, there is a description of proposed architectures and description of the experiment. In the last part of the work, multiframe methods and single frame methods are compared. In the result, the proposed methods recover more details, however, some architectures produce artefacts, which can be reduced using a filter, for example, Gaussian. New methods allow to reduce the number of failed face recognition. This fact is necessary for person identification in case of incidents.
|
|
|
Protection of sensitive data contained in images
Mezina, Anzhelika ; Rajnoha, Martin (oponent) ; Burget, Radim (vedoucí práce)
This work is focused on application of deep learning in security problem of escape sensitive information, that is contained in images. The presented solution of this problem is using Single Shot Multibox Detector and Fully Connected Network (FCN). FCN is faster than other methods and can be applied in industry, where is a need to analyse input and output information very quickly, for example, in network traffic analysis. In the first part of this work, methods that can be used in keyword detection are described. The second part contains a description of experiment and achieved results for two models of neural network: Single Shot Multibox Detector and Fully Connected Network. The second one gave better results and can be used in practice.
|
host ::
RSS.