|
|
Color-to-Grayscale Video Conversions
Března, Filip Samuel ; Hradiš, Michal (referee) ; Čadík, Martin (advisor)
Color to grayscale conversion is still a relevant topic and finds use in multiple areas, to which belongs artistic photography, but primarily colorless print and for simplifying some processes in picture processing. This Bachelor thesis deals with this conversion - decolorization, with focus on video sequences. The basic principle of representation of digital image is explained here alongside with operations for its processing used in next part of thesis. Thereafter it includes the analysis of types of methods, which are used for the conversion to grayscale and further their properties and success rate are evaluated aswell. In the second part of this thesis, there are three chosen and implemented decolorization methods and summary of their achieved results, which is then accompanied by evaluation of practicability in conversion of colorful video sequences to grayscale ones.
|
|
|
Realistic Landscape with Vegetation
Zelený, Jan ; Jošth, Radovan (referee) ; Hradiš, Michal (advisor)
There is enough rendering power to draw more than only simple indoor scenes today and it can produce very realistic images of landscape with vegetation. Moreover, there are new sophisticated methods for generating of such landscape and simulation of plants ecosystem. This text explains few algorithms for generating and methods for interactive rendering of landscape and vegetation.
|
|
|
Learning to Generate Images with Convolutional Neural Networks
Kohút, Jan ; Kolář, Martin (referee) ; Hradiš, Michal (advisor)
The aim of this Bachelor's thesis is to design and analyze convolutional neural networks generating images of characters based on their parameters. Parameters of characters are type of char, font, colour of character, background colour, translation and rotation. Neural networks have created multidimensional representation of each parameter. Relations inside these representation are similar to relations inside parameters. Neural networks generate characters with new values of parameters based on interpolation between learned values of parameters. Neural networks are capable to generalize problem of generating images.
|
|
|
Convolutional Networks for Historic Text Recognition
Macurová, Nela ; Špaňhel, Jakub (referee) ; Hradiš, Michal (advisor)
This thesis deals with the recognition of historical texts using deep neural networks, specifically the recognition of individual words in Gothic script in Czech. Here is a general overview of convolutional networks and text recognition methods. A dataset was created with real and generated data. The network was trained on generated data and testing on real images of words. This proposed word classification method was not very successful due to different test and training data.
|
|
|
Pedestrian Identification
Jurča, Jan ; Špaňhel, Jakub (referee) ; Hradiš, Michal (advisor)
This thesis deals with pedestrian identification from video sequence based on person, face and gait recognition. For person and face recognition are used pretrained networks. While for gait recognition is implemented and compared many different networks. Final pedestrian recognition is based on multimodal fusion realized by neural network. For the purpose of the work was created dataset, along with a set of tools that allow its almost automatic creation.
|
|
|
Deep Learning for Facial Recognition in Video
Mihalčin, Tomáš ; Sochor, Jakub (referee) ; Hradiš, Michal (advisor)
This diploma thesis focuses on a face recognition from a video, specifically how to aggregate feature vectors into a single discriminatory vector also called a template. It examines the issue of the extremely angled faces with respect to the accuracy of the verification. Also compares the relationship between templates made from vectors extracted from video frames and vectors from photos. Suggested hypothesis is tested by two deep convolutional neural networks, namely the well-known VGG-16 network model and a model called Fingera provided by company Innovatrics. Several experiments were carried out in the course of the work and the results of which confirm the success of proposed technique. As an accuracy metric was chosen the ROC curve. For work with neural networks was used framework Caffe.
|
|
|
Active Learning with Neural Networks
Bureš, Tomáš ; Kolář, Martin (referee) ; Hradiš, Michal (advisor)
The topic of this thesis in active learning in conjunction with neural networks. First, it deals with theory of active learning and strategies used in real life scenarios. Followed by practical part, experimenting with active learning strategie and evaluating those experiments.
|
|
|
Automatic Finish Camera
Jahoda, Vojtěch ; Zemčík, Pavel (referee) ; Hradiš, Michal (advisor)
This thesis addresses the automation of photo finish evaluation in athletics. The detection of athletes in the target photo has been performed using the OpenPose library. Subsequently, athlete background segmentation has been performed to remove noise and cropping for individual athletes. The evaluation itself has been solved using a regression convolutional neural network. The accuracy of 77.60% has been achieved trough detecting the figures in the photo and 89.28% of the evaluated records has reached the accuracy lesser than 10 ms. The main benefit of this thesis is for novice photo finish referees, since they will have the evaluated target record available to them beforehand. Another usage serves for coaches and competitors, since they will be able to easily validate the evaluated records by themselves.
|
|
|
Improving Consistency in Text Recognition Datasets
Tvarožný, Matúš ; Hradiš, Michal (referee) ; Kišš, Martin (advisor)
This work is concerned with increasing the consistency of datasets for text recognition. This paper describes the problems that cause the inconsistency and then presents solutions to eliminate it. The effect of the properties of the polygons defining the text line boundaries and hence how the modified version of the dataset, which is composed of ideal text line variants, affected the accuracy of the model is investigated. Further, the work focuses on detecting and then removing or modifying text lines whose ground truth transcription does not match the actual text they contain. Experimentation showed that removing the visual inconsistency on the training set did not have a significant effect on the trained model, but modifying the test set improved the OCR accuracy of the model by 1.1\% CER. By modifying the dataset so that it did not contain mutually inconsistent pairs of recognized text and the corresponding ground truth, the model improved by a maximum of only 0.2\% CER after re-training. The main finding of this work is, above all, the proven beneficial effect of removing inconsistencies on test suites, thanks to which it is possible to determine a more realistic error rate of the OCR model.
|
|
|
Music Style Recognition
Behúň, Kamil ; Polok, Lukáš (referee) ; Hradiš, Michal (advisor)
This thesis deals with the music style recognition. The introduction is an overview of current methods used in the music style recognition. Next chapters deals with the system created for the music style recognition. The final system is consists of two feature extraction methods. The first uses the Mel-frequency cepstral coefficients extraction from records and the second uses feature extraction from spectrograms of records. The final system uses Support Vector Machine for classifying.
|
guest ::
