Evaluating Non-Reference Image Quality Metrics for AI-Generated Images: A Novel Approach

This research paper explores the development and evaluation of non-reference image quality metrics specifically tailored for AI-generated images created by task-specific prompts. Given the unique challenges posed by such images, traditional metrics often fall short in assessing their perceptual quality and alignment with the provided prompts. This study introduces a novel approach that integrates multi-granularity similarity measurements and task-specific prompts to evaluate both perceptual and alignment quality. Experimental results on benchmark datasets demonstrate the effectiveness of the proposed metrics, offering a new standard for assessing AI-generated images.