Photorealistic stylised image quality assessment database (PSIQAD) building and modelling

• Main Author: Low, Qing Ru
• Other Authors: Lin Weisi
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2020
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence; Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision
• Online Access: https://hdl.handle.net/10356/138132
• Institution: Nanyang Technological University

Image Quality Assessment (IQA) tasks have increasing importance in today’s context due to the ubiquitous use of imaging devices and image-editing applications. Despite having several existing IQA models, they usually evaluate the degradation or aesthetic aspect of an image. The emergence of Partially Artificial Images (PAIs), whose contents are partially or completely generated by image generation algorithms [39], brings more challenges to the applicability of conventional IQA methods since both enhancements and distortions exist in the generation process of PAIs. This project also discusses why conventional IQA metrics are unable to work for PAIs which mainly lies in the knowledge fed to build IQA metrics. A novel image database, Photorealistic Stylised Image Quality Assessment Database (PSIQAD), is introduced to analyse the human preference in photorealistic stylised images, a form of PAIs, with the creation of baseline objective Stylised IQA (SIQA) models to show how PSIQAD can be leveraged. The advantages of PSIQAD with respect to the existing databases were also reviewed.