No-Reference Image Quality Assessment Method Based on High-Level Semantic Features

Image Quality Assessment (IQA) aims to estimate the perceptual quality of images by simulating the human visual system. However, existing no-reference IQA methods often fail to fully exploit high-level semantic information. To address this limitation, we propose a novel no-reference image quality assessment method based on high-level semantic features. Specifically, the structural and color information of the image is first hierarchically fused. Then, an adaptive gating mechanism is introduced to dynamically adjust the feature retention strategy when computing the correlation between low-level and high-level features. Finally, high-level semantic information is guided toward low-level features to help the network focus on semantically important distorted regions. Experimental results demonstrate that the proposed method significantly enhances the model’s global perceptual capability through the collaborative learning of high-level semantic and low-level features. It also exhibits stronger robustness under complex distortion conditions.