Baqar, Mohtashim * (2024) Towards a unified image quality assessment technique for cross-content image processing applications. Doctoral thesis, Sunway University.
Full text not available from this repository.Abstract
The rise in consumer electronics has led to the daily generation of massive multimedia signals. To make efficient multimedia processing systems, understanding human perception is crucial. This has sparked significant interest in image quality assessment (IQA) and its applications within the research community. This study focuses on creating IQA tools that align with human perception, are practical, and can enhance multimedia content. It addresses two key issues: unified image quality assessment for diverse content and perceptually-driven image reconstruction. The study aims to address gaps in image quality research, particularly the lack of suitable IQA metrics for cross-content multimedia applications. Hence, the issue of cross-content type, i.e., screen-content images (SCIs) and natural-scene images (NSIs), image quality assessment is the main objective of this study. The study’s second objective is to apply IQA practically in image processing applications/operations. It specifically tackles image reconstruction, a vital post-processing task in image systems. Most state-of-the-art (SOTA) image reconstruction methods areapplication-specific (AS), requiring extra information like tuning parameters to correct distortions. However, practical scenarios often involve unknown causes of distortion, making reconstruction complex. While some generic reconstruction methods exist, they lag AS ones in performance. Additionally, these methods often don’t consider visual perception, even though human vision is crucial in most applications. The study aims to develop methods for reconstructing distorted images while prioritizing visually important areas. This leads to the development of a two-stage IQA-based perceptual image refinement framework, acknowledging the challenge of translating IQA metrics into practical applications. In summary, the study develops algorithms for accurate image quality prediction, refining images, and enhancing visual quality. For the first objective, two metric types are created: Permutation Entropy-Based (PEFRF, IW-PEFRF) and Deep Feature Similarity (DFeaSim) for better image quality assessment. For the second objective, the study introduces a two-stage perceptually-driven image reconstruction framework. It can reconstruct images with five distortion types, outperforming SOTA AS algorithms, even under high distortion. The first stage uses an application-specific reconstruction algorithm, while the second stage employs an IQA-based model called the observation-based bilateral filter (OBF) with non-linear weights calculated using a Haar-PSI-based maximum a posteriori (MAP) estimator. Further, the study utilized 13 benchmark IQA databases with diverse image contexts, appearances, and quality distortions to develop and validate the proposed models. Experimental results demonstrated the robustness of the quality assessment and refinement models compared to existing techniques. PEFRF and IW-PEFRF IQA methods achieved significantly higher performance indexes (PLCC, SRCC, KRCC, RMSE) than classical and SOTA techniques, with values of 0.9644, 0.9687, 0.8419, and 0.0585 for PEFRF, and 0.9647, 0.9690, 0.8420, and 0.0584 for IW-PEFRF, averaged over 20,000 images. Subsequently, the HaarPSI-based reconstruction approach improved accuracy from 2% to 84% compared to AS algorithms at both low and high distortion levels. There is no doubt that the findings put forward in the study will aid in the development of perceptually driven image processing systems and applications in the future.
| Item Type: | Thesis (Doctoral) |
|---|---|
| Uncontrolled Keywords: | cross-content type applications; unified image quality assessment; perceptual refinement; image reconstruction. |
| Subjects: | T Technology > TA Engineering (General). Civil engineering (General) |
| Divisions: | Sunway University > School of Engineering and Technology [formerly School of Science and Technology until 2020] > Dept. Computing and Information Systems |
| Depositing User: | Ms Yong Yee Chan |
| Date Deposited: | 29 Jul 2025 05:38 |
| Last Modified: | 29 Jul 2025 05:38 |
| URI: | http://eprints.sunway.edu.my/id/eprint/3242 |
Actions (login required)
 |
View Item |
