Perceptual Quality Assessment for AI Image Enhancement

Perceptual Quality Assessment for AI Image Enhancement

With the rapid advancement of AI-powered image enhancement techniques, methods such as super-resolution, denoising, color enhancement, and style transfer have been widely applied in fields like photography and multimedia content creation. However, objectively and accurately assessing the perceptual quality of AI-enhanced images remains a challenging problem. Traditional image quality assessment (IQA) metrics, such as PSNR and SSIM, often fail to adequately capture the perceptual experience of the human visual system. Focusing on perceptual quality assessment for AI image enhancement, this talk provides a systematic overview of mainstream full-reference and no-reference IQA methods, discusses recent advances in deep learning-based IQA models, analyzes common distortion patterns found in AI-enhanced images, and proposes a new IQA paradigm specifically designed for AI-enhanced images.