Text this: Enhancing full-reference image quality assessment via dual-attention fusion of global–local features and CLIP-based text prior.