|
| |
|
|
Detect the Use of Retouch Filters in a Face Image
Kraváček, Adam ; Drahanský, Martin (referee) ; Goldmann, Tomáš (advisor)
These days, altering images via filters is one of the easiest ways of enhancing its properties. Social networks like Instagram or Snapchat, focused primarily on image sharing, offer their users the option to apply filters on their images, which alter their colours to make them look better. If someone was to extract images from these platforms, many of these images would have a filter applied. This thesis explains the principles of these filters and focuses on detection of filters on facial images. Several approaches to detecting filters are being experimented with. Detection by analysis of histograms and detection by convolutional neural network achieve the best results and so are implemented in a program with a simple user interface. They achieved a success rate of 94,44% (histogram) and 99,10% (CNN). This thesis also investigates the impact of filters on facial recognition, where the impact varies depending on the filter used. Some filters have a significant impact on the rate of successful identifications, whereas others have little impact.In general, however, it can be said that the changes introduced by the application of filters are not negligible.
|
|
|
Typing Using Brain Signals
Wagner, Lukáš ; Malinka, Kamil (referee) ; Tinka, Jan (advisor)
This bachelor thesis focusses on the implementation of a brain-computer interface, programmed in Python language, that would enable to communicate using EEG. The thesis investigates and evaluates existing brain-computer interface technologies for this purpose. The thesis also explores the use of machine learning applied to the technology, in particular neural networks, which have proven to be one of the most accurate methods of EEG signal processing. Following that, 3 different systems are proposed and implemented, each on different paradigm of visually evoking EEG potential changes. These systems were tested with different signal classification approaches. Unfortunately, none of the systems proved to be useful in communication.
|
|
|
Visual Localization of Chess Pieces
Hampl, Tomáš ; Špaňhel, Jakub (referee) ; Hradiš, Michal (advisor)
The main goal of this thesis was to analyze state of the chess game and to locate chess pieces on the chessboard. Chessboard recognition is based on locating lines in image using Hough transform and PClines. The figures were detected by models of convolutional neural networks - YOLOv3, YOLOv4 and YOLOv4 tiny. Evaluation was perfomed on our data set. Chessboard detection achieves accuracy of 97%, complete localization of the state reaches 74,5% and piece localization 96%.
|
|
|
Depth-Based Determination of a 3D Hand Position
Ondris, Ladislav ; Tinka, Jan (referee) ; Drahanský, Martin (advisor)
Cílem této práce je určení kostry ruky z hloubkového obrazu a jeho následné využití k rozpoznání statického gesta. Na vstupu je hloubkový obrázek, ve kterém je nejprve detekována ruka pomocí neuronové sítě Tiny YOLOv3. Následně je obrázek zbaven pozadí a z takto předzpracovaného obrázku je určena kostra ruky v podobě 21 klíčových bodů neuronovou sítí JGR-P2O. K rozpoznání gesta z klíčových bodů ruky byla navržena technika, která porovná kostru na vstupu s uživatelem definovanými gesty. Funkcionalita systému byla otestována na vytvořeném datasetu s více než čtyřmi tisíci obrázky.
|
|
|
PVC detection in ECG
Imramovská, Klára ; Hejč, Jakub (referee) ; Ronzhina, Marina (advisor)
The thesis deals with problems of automatic detection of premature ventricular contractions in ECG records. One detection method which uses a convolutional neural network and LSTM units is implemented in the Python language. Cardiac cycles extracted from one-lead ECG were used for detection. F1 score for binary classification (PVC and normal beat) on the test dataset reached 96,41 % and 81,76 % for three-class classification (PVC, normal beat and other arrhythmias). Lastly, the accuracy of the classification is evaluated and discussed, the achieved results for binary classification are comparable to the results of methods described in different papers.
|
|
|
Object detection in video using neural networks and Android application
Mikulec, Vojtěch ; Kiac, Martin (referee) ; Myška, Vojtěch (advisor)
This master’s thesis deals with the implementation of functional solution for classifying road users using mobile device with Android operating system. The goal is to create Android application which classifies vehicles in real time using rear-facing camera and saves timestamps of classification. Testing is performed mostly with own, diversely modificated dataset. Five models are trained and their performance is measured in dependence on hardware. The best classification performance is from pretrained MobileNet model where transfer learning with 6 classes of own dataset is used – 62,33 %. The results are summarized and a method for faster and more accurate traffic analysis is proposed.
|
|
|
Reconstruction of Missing Parts of the Face Using Neural Network
Marek, Jan ; Drahanský, Martin (referee) ; Goldmann, Tomáš (advisor)
Cílem této práce je vytvořit neuronovou síť která bude schopna rekonstruovat obličeje z fotografií na kterých je část obličeje překrytá maskou. Jsou prezentovány koncepty využívané při vývoji konvolučních neuronových sítí a generativních kompetitivních sítí. Dále jsou popsány koncepty používané v neuronových sítích specificky pro rekonstrukci fotografií obličejů. Je představen model generativní kompetitivní sítě využívající kombinaci hrazených konvolučních vrstev a víceškálových bloků schopný realisticky doplnit oblasti obličeje zakryté maskou.
|
|
|
Time-Domain Neural Network Based Speaker Separation
Peška, Jiří ; Černocký, Jan (referee) ; Žmolíková, Kateřina (advisor)
A thesis is about the usage of convolutional neural networks for automatic speech separation in an acoustic environment. The goal is to implement the neural network by following a TasNet architecture in the PyTorch framework, train this network with various values of hyper-parameters, and to compare the quality of separations based on the size of the network. In contrast to older architectures that transformed an input mixture into a time-frequency representation, this architecture uses a convolutional autoencoder, which transforms input mixture into a non-negative representation optimized for a speaker extraction. Separation is achieved by applying the masks, which are estimated in the separation module. This module consists of stacked convolutional blocks with increasing dilation, which helps with modeling of the long-term time dependencies in processed speech. Evaluation of the precision of the network is measured by a signal to distortion (SDR) metric, by a perceptual evaluation of speech quality (PESQ), and the short-time objective intelligibility (STOI). The Wall Street Journal dataset (WSJ0) has been used for training and evaluation. Trained models with various values of hyper-parameters enable us to observe the dependency between the size of the network and SDR value. While smaller network after 60 epochs of training reached 10.8 dB of accuracy, a bigger network reached 12.71 dB.
|
|
|
Impact of color models on performance of convolutional neural networks
Šimunský, Martin ; Doležel, Petr (referee) ; Škrabánek, Pavel (advisor)
Current knowledge about impact of colour models on performance of convolutional neural network is investigated in the first part of this thesis. The experiment based on obtained knowledge is conducted in the second part. Six colour models HSV, CIE 1931 XYZ, CIE 1976 L*a*b*, YIQ a YCbCr and deep convolutional neural network ResNet-101 are used. RGB colour model achieved the highest classification accuracy, whereas HSV color model has the lowest accuracy in this experiment.
|
guest ::
