The combined use of statistical and deep learning methods in the removal of periodic noise

In this thesis, a two-stage solution has been developed using deep learning methods to remove periodic noise frequently encountered in digital images. The study emphasizes that periodic noise is a common problem in digital imaging systems and that this noise significantly degrades image quality, making analysis and processing difficult. The aim of the study is to provide a more effective noise reduction solution using deep learning techniques in situations where traditional methods are insufficient. In this context, a two-stage model has been developed by combining the DnCNN (Denoising Convolutional Neural Network) model with frequency domain filtering methods. The performance of this model has been evaluated on different types of noise and various image datasets, yielding superior results compared to traditional methods. The thesis first thoroughly examines the definition and types of noise, followed by an explanation of the fundamental principles and application areas of deep learning methods. In conclusion, it has been demonstrated that the developed two-stage model effectively removes periodic noise and significantly improves image quality.