华中科技大学学报（自然科学版）2025，Vol.53Issue(5)：71-77,127,8.DOI:10.13245/j.hust.250104
一种补偿互联网传输降质的人脸图像增强方法
A face image enhancement method to compensate for degradation of Internet transmission
摘要
Abstract
Based on skip-connection shifted window self-attention mechanism,an enhancement method with low time complexity was proposed to enhance degraded face images transmitted by Internet applications.In this method,global features of image were effectively extracted by integrating window self-attention mechanism and shifted window self-attention mechanism,and some complex self-attention calculations with high similarity were omitted by skip connection.Taking WeChat software transmission degradation as an example,based on the existing CelebAMask dataset,a shareable dataset of degraded face images for Internet transmission was fabricated.Multi-parameter ablation experiments and multi-method comparison experiments were carried out on this dataset to verify the effectiveness of the proposed method.Experimental results show that compared with other current face image enhancement methods,the proposed method can significantly reduce the time complexity while maintaining the subjective visual effect and objective index value of the reconstructed image.关键词
自注意力机制/互联网传输降质/人脸图像增强/跳跃连接/重构图像/时间复杂度Key words
self-attention mechanism/degradation of Internet transmission/face image enhancement/skip-connection/reconstructed image/time complexity分类
信息技术与安全科学引用本文复制引用
肖然,刘昱,刘人赫,王瀚..一种补偿互联网传输降质的人脸图像增强方法[J].华中科技大学学报（自然科学版）,2025,53(5):71-77,127,8.基金项目
天津市自然科学基金资助项目(22JCZDJC00270) （22JCZDJC00270）
天津市研究生科研创新项目(2021YJSO2B12). （2021YJSO2B12）
