Image Quality Assessment Characteristics for Super-Resolution

With the rapid advancement of artificial intelligence (AI), Image Quality Assessment (IQA) techniques are increasingly used to evaluate the quality of images reconstructed through Super Resolution (SR). However, most IQA methods do not consider how the human eye assesses image quality; they primarily rely on complex networks for accuracy. This paper examines the relationship between humans and AI, arguing that IQA methods that integrate both top-down and bottom-up processes better align with human evaluation, whereas AI typically uses a solely bottom-up approach. We evaluate images generated by various SR algorithms using these two IQA methodologies. Experimental results show that humans prefer SR algorithms that emphasize global image features, while AI favors those focusing on local features. These findings highlight a divergence in image quality preferences between humans and AI, underscoring the need to select algorithms based on specific application scenarios.