Scholarlink

<p>Existing blind image quality assessment (BIQA) networks cannot well simulate human visual system (HVS) and have poor subjective and objective consistency in IQA. To solve such problems, a new BIQA network is proposed based on 2D Log-Gabor filter and ResNet50, named as GFNET. 2D Log-Gabor filter is used to simulate HVS to extract significant feature of the input images. The ReLU in ResNet50 is replaced by Better-Softplus which is proposed in this paper. The full connected layer of ResNet50 is replaced by a regression module. Sigmoid function is used to map the network output to [0,1] to get the final score. Spearman rank-order correlation coefficient (SROCC) and Pearson linear correlation coefficient (PLCC) are used to evaluate the proposed network and compare its performance with existing BIQA networks on typical datasets. Experimental results show that the proposed GFNET network has good subjective and objective consistency. The average values of PLCC and SROCC reached 0.892 and 0.878, respectively. It can well simulate HVS and accurately reflect the visual perception effect of HVS on image quality.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

GFNET: A blind image quality assessment network based on gabor filter and ResNet50

  • 期刊论文

  • 发表日期：

    2026年5月25日

查看全文
  • Hao-Ran Sun,
  • Tao Li,
  • Jin-Rong Li,
  • Bao-Jun Shi

摘要

Existing blind image quality assessment (BIQA) networks cannot well simulate human visual system (HVS) and have poor subjective and objective consistency in IQA. To solve such problems, a new BIQA network is proposed based on 2D Log-Gabor filter and ResNet50, named as GFNET. 2D Log-Gabor filter is used to simulate HVS to extract significant feature of the input images. The ReLU in ResNet50 is replaced by Better-Softplus which is proposed in this paper. The full connected layer of ResNet50 is replaced by a regression module. Sigmoid function is used to map the network output to [0,1] to get the final score. Spearman rank-order correlation coefficient (SROCC) and Pearson linear correlation coefficient (PLCC) are used to evaluate the proposed network and compare its performance with existing BIQA networks on typical datasets. Experimental results show that the proposed GFNET network has good subjective and objective consistency. The average values of PLCC and SROCC reached 0.892 and 0.878, respectively. It can well simulate HVS and accurately reflect the visual perception effect of HVS on image quality.