State of the Art in Super-Resolution: A 2024 Comparison of Techniques and Models

Abstract

This is a comprehensive comparison study of contemporary super-resolution (SR) techniques and models as of 2024. This Thesis highlights the significant advancements in computer vision and image processing. Also, It include the Traditional methods such as interpolation based and frequency domain approaches. With the introduction of deep learning, particularly the Convolutional Neural Networks (CNNs) which transformed the SR landscape by automating complex feature extraction and enabling more accurate image reconstruction. Key models discussed in this dissertation are SRCNN (which introduced end to end mapping for SR), VDSR and FSRCNN. This Thesis also explores advancements in real time applications with ESPCN and the multi scale capabilities of LapSRN.