差异化建模与感知增强的轻微失真盲图像质量评价OA

Despite significant progress in mainstream blind image quality assessment(BIQA)algorithms,performance bottlenecks persist when evaluating slightly distorted images.This stems from feature modeling conflicts induced by global training strategies:the human visual system(HVS)focuses on global degradation for severe distortions but relies heavily on local discrepancies and semantic content for slight ones.To address this,a BIQA framework was proposed based on differential modeling(DM)and perception enhancement.Firstly,a DM strategy constructed a dual-branch ar-chitecture to decouple slight and severe distortion representations,mitigating cross-level conflicts.Secondly,a semantic guidance perception enhancement(SGPE)module adaptively directed the model to focus on semantically sensitive re-gions,enhancing sensitivity to subtle quality variations.Finally,a distortion-adaptive weighted fusion(DAWF)strategy dynamically aggregated features,significantly improving slight-distortion accuracy while maintaining robust prediction across the full distortion spectrum.Experiments on standard benchmark datasets demonstrate that the proposed method effectively overcomes the slight-distortion bottleneck and exhibits strong generalization capability.