Development of Efficient Image Super Resolution Algorithms using Hybrid Domain Feature Extraction Principles

Abstract

An image super-resolution creates a high-resolution image from low-resolution images for better image visualization for several recent applications. Advances in super-resolution were introduced three decades ago, and considerable improvements were made in computing power using magnitude, digital cameras, and detailed digital displays. With technological advancement, the viewer s expectations have also increased for super-resolution images and videos. Every super-resolution algorithm must undergo three steps, one is the need for accurate motion estimation, the second is the de-blurring, and the last step is the robustness for stochastic errors and modeling. Chapter 1 deals with the Super Resolution (SR), types of SR s, applications, motivation, and research objectives. A detailed literature survey of SR based on spatial, transform domain, and Deep learning techniques presented by several researchers is explained in chapter 2. newlineAn image super-resolution based on the fusion of the novel Average Pixel Values Technique (APVT) and Discrete Wavelet Transform (DWT) is proposed in chapter 3. The Low Resolution (LR) images are considered and converted into High Resolution (HR) images using a novel technique of APVT by inserting an average of rows between rows and an average of columns between columns to get HR images. The DWT is applied to HR images to obtain four bands. The HR image quality is enhanced by Histogram Equalization (HE). The LL band reference image and HE matrix are added to obtain a new LL band. The inverse DWT is applied on four bands to derive the Super Resolution (SR) image. newlineIn chapter 4, an average segmentation-based image super-resolution using wavelet coefficients augmentation is projected. The Discrete Wavelet Transform (DWT) is applied on a reference image of 166x304 and obtained four bands with each size of 83x152. The LR images are obtained by combining four bands and dividing each coefficient by two. The HR images are obtained by inflating one coefficient into 2x2 coefficients of LR images. The 3x3