Ph.D. Tezi Görüntüleme

      In the thesis work, novel methods are proposed to increase the spatial resolution of a low resolution image which the process is referred to as super resolution image reconstruction in the literature. The first contribution to the literature is to present a novel edge, corner and boundary preserving sparse representation based super resolution approach by using Wavelet transform. The second contribution of the thesis to the literature is the use of Gabor Wavelets method fort he purpose of feature extraction. The experiments conducted on synthetic images show that the proposed method gives better results compared to learning-based method except for deep learning in the literature. The third contribution of the thesis to the literature is to analyze the convolutional neural network based super resolution approaches in the literature and to propose a novel convolutional neural network structure with deep skip-connections for learning-based super-resolution. With this network structure, the super resolution qualities of synthetic images have been improved. As the fourth contribution of the thesis to the literature, the microscopic image super resolution is examined with a novel convolutional neural network structure. The proposed study eliminates the limitations of the first and only study that provides the super resolution of microscopic images in the literature. Finally, high quality images are obtained with the proposed sparse representation based pan sharpening approach in remote sensing images and a detailed comparison is made for the first time in the literature.