基于局部特征非均匀提取的全景图像质量评价OA

The existing approaches for evaluating the quality of panoramic images suffer from issues such as lack of consistency between the observation space and representation space,and limited ability to fuse local quality.To address these issues,this paper proposes a quality assessment approach for panoramic images by leveraging non-uniform local feature extraction.The proposed method systematically explored the characteristics of both the human visual system and panoramic images,and introduced a fine-grained strategy for non-uniform local feature extraction in the feature domain.Moreover,it also utilized a self-attention mechanism to consider the quality dependencies among regions through a global perspective,which effectively aggregated the local perceptual features of each region.The experimental results on the CVIQ,LIVE 3D VR and MAVQD datasets validate the accuracy of the proposed framework and demonstrate the effectiveness of each module involved.