|
|
Rotation-invariant pattern matching for egg recognition
Bláha, Vojtěch ; Vomlelová, Marta (advisor) ; Harmanec, Adam (referee)
This bachelor's thesis follows up the recognition of eggs in the image. The goal was to create a group of programs that firstly captures image data, than finds eggs in them and finally shows the results in some user environment. We tested gradually different classification methods (template matching, logistic regression and neural network). We tried also different representations of the image such as matrix representation and ring projection, and various pre-processing of the image before the actual finding, we used grayscale, color spectra and edges detected by a high-pass or Kirsche detector. After testing all methods, we selected the best one and created the classification program itself. The most successful method was logistic regression with ring projection. 1
|
|
|
Registration of images of nuclear fuel assembly
Harmanec, Adam ; Blažek, Jan (advisor) ; Šikudová, Elena (referee)
Nuclear fuel is visually inspected during regular shutdowns in order to monitor defects and long-term changes. To enable automatic comparison of images of fuel assemblies, it is crucial to perform their registration, the implementation of which has not yet been published in the scientific literature. In this work we present an analysis of image registration techniques and similarity metrics inspired by the focus operators used in autofocus and shape-from-focus. Their performance has been evaluated using a series of experiments that tested their various properties on a novel data set obtained in cooperation with the research organization Centrum výzkumu Řež. Finally, we present and discuss the results and make recommendations on which to use in which scenario.
|
|
|
Traffic sign classification by deep learning
Harmanec, Adam ; Blažek, Jan (advisor) ; Kratochvíl, Miroslav (referee)
Classification of road signs has been studied for many years and very promising results have been achieved. We present the analysis of used data sets as very limited for real case classification. In this thesis we analyse publicly available data sets and by merging and extending them, we create a wider and more comprehensive data set applicable in the Czech Republic. Finally, we propose a new convolutional neural network architecture and test it along with several preprocessing techniques on the new data set reaching accuracy of over 99%.
|
guest ::
