APPLICATION OF DEEP LEARNING ALGORITHM ENHANCED SUPER RESOLUTION GENERATIVE ADVERSARIAL NETWORK (ESRGAN) BASED ON SINGLE IMAGE SUPER RESOLUTION USING WEBSITE USER INTERFACE

• Main Author: Angelina Samosir, Widia
• Format: Final Project
• Language: Indonesia
• Subjects: Teknik (Rekayasa, enjinering dan kegiatan berkaitan)
• Online Access: https://digilib.itb.ac.id/gdl/view/48020
• Institution: Institut Teknologi Bandung

High-resolution images have become a fundamental requirement in this era. This is because the demand for high-quality content is getting increase every year. However, there are still difficulties in processing images that have low resolution to be used as high-resolution images. This is because the existing image processing system, it is still difficult to make improvements to low-resolution images easily and quickly. Therefore, we need a system that can make image repairs quickly, easily, and can be used by everyone. Single Image Super-Resolution (SISR) is a technology capable of producing high-resolution images with realistic textures from low-resolution images by inserting a single image into the model. The aim of our research is to build a deep learning model for Single Image Super-Resolution (SISR) and to integrate the deep learning model into the user interface. So that everyone can use the model easier and faster. In this Final Project, the author designs a system that can implement SISR technology into a User Interface. This system uses an Enhanced Super-Resolution Generative Adversarial Network (ESRGAN) with a block of Residual-in-Residual Dense blocks (RRDB) in system design. The user interface of the model is made in the form of a website. Making back-end uses Flask to deploy ESRGAN deep learning models. In building front-end, HTML and CSS are used to design the user interface. This system can build better visual quality with realistic and natural textures from high-resolution images from low-resolution images. The benefits of this project are applied deep learning model using website based on SISR to produce high-quality results that can be processed with all types of the low-resolution input images (people, animals, flowers, etc.) and it is easier to get results using the website pages that have been provided.