Beyond Pixel: Why Image Quality Is More Than a Number?
A research presentation at Melbourne Institute of Technology's two-day Research Seminar, held on 27–28 October 2026.
Looking beyond pixel-level error
Image quality is not simply about how closely two images match at the pixel level. In this presentation, I discussed how different kinds of content—particularly natural-scene and screen-content images—create different challenges for visual quality assessment. The talk connected my research on image quality measures with perceptual image reconstruction and restoration.
It also explored the tension between making an image look better and preserving information in the original scene, a question that has become more important with generative AI-based restoration.
Topics discussed
- Natural-scene versus screen-content image quality, including cross-content assessment.
- Permutation-entropy and learning-based approaches to image quality assessment.
- HaarPSI-based refinement in perceptual image reconstruction, with comparisons using PSNR, SSIM and HaarPSI.
- Examples from underwater, medical and satellite imaging, discussed as wider applications rather than experimentally tested domains of the presented study.
- Why perceptual realism and the faithful preservation of meaningful information both matter in generative AI image restoration.
View the research seminar slides
The complete 13-slide presentation is available below as a PDF. You can view it here, open it in a separate tab, or download a copy.
If the slides do not appear in your browser, use the direct PDF link.