Redukcia tieňov v obraze

Abstract

The diploma thesis adresses the issues of reducing shadows in an image. The aim of the thesis is to create an application for the detection and reduction of shadows using convolution neural networks. For the creation of the application, I chose the programming language Python and the framework for the machine learning TensorFlow. In the introductory part of the thesis, I describe techniques to reduce shadows, both without and with machine learning. The study further includes a description of networks for generating images of Generative Adversarial Networks (GAN). Subsequently, in the next chapter I am concerned with a detailed procedure of the practical part, which includes an implementation of the selected network Stacked Conditional Generative Adversarial Network (ST-CGAN). In conclusion, I evaluate the achievements and personal benefits of the study.