DGSF-AOT：动态门控与自注意力融合增强的人脸图像修复

柏武贰; 张乾; 刘霜; 滕林; 杨思红

doi:10.11959/j.issn.1000-0801.2026016

您当前的位置：

首页 >

文章列表页 >

DGSF-AOT：动态门控与自注意力融合增强的人脸图像修复

研究与开发 | 更新时间：2026-03-05

- DGSF-AOT：动态门控与自注意力融合增强的人脸图像修复
- DGSF-AOT: dynamic gating and self-attention fusion enhancement for face image restoration
- 电信科学 2026年42卷第2期页码：120-134
- 作者机构：
  
  1.贵州民族大学数据科学与信息工程学院，贵州贵阳 550025
  2.贵州省模式识别与智能系统重点实验室，贵州贵阳 550025
  3.贵州民族大学教务处，贵州贵阳 550025
- 作者简介：
  
  [ "柏武贰（1995- ），男，贵州民族大学数据科学与信息工程学院硕士生，主要研究方向为图像修复。" ]
  [ "张乾（1984- ），男，博士，贵州民族大学教务处教授，主要研究方向为机器学习、模式识别、计算机视觉。" ]
  [ "刘霜（2001- ），女，贵州民族大学数据科学与信息工程学院硕士生，主要研究方向为图像修复。" ]
  [ "滕林（1996- ），男，贵州民族大学数据科学与信息工程学院硕士生，主要研究方向为图像修复。" ]
  [ "杨思红（1996- ），女，贵州民族大学数据科学与信息工程学院硕士生，主要研究方向为图像修复。" ]
- 基金信息：
  
  贵州省高等学校大数据分析与智能计算重点实验室项目(黔教技[2023]012号);贵州民族大学校级科研项目(GZMUZK [2021] YB23;GZMUZK [2023] QN10)
- DOI：10.11959/j.issn.1000-0801.2026016
  中图分类号： TN957.52;TP391.41
- 收稿：2025-06-09，
  
  修回：2025-07-04，
  
  录用：2025-08-29，
  
  纸质出版：2026-02-20
- 稿件说明：
移动端阅览
柏武贰,张乾,刘霜等.DGSF-AOT：动态门控与自注意力融合增强的人脸图像修复[J].电信科学,2026,42(02):120-134.

Bai Wuer,Zhang Qian,Liu Shuang,et al.DGSF-AOT: dynamic gating and self-attention fusion enhancement for face image restoration[J].Telecommunications Science,2026,42(02):120-134.
柏武贰,张乾,刘霜等.DGSF-AOT：动态门控与自注意力融合增强的人脸图像修复[J].电信科学,2026,42(02):120-134. DOI： 10.11959/j.issn.1000-0801.2026016.

Bai Wuer,Zhang Qian,Liu Shuang,et al.DGSF-AOT: dynamic gating and self-attention fusion enhancement for face image restoration[J].Telecommunications Science,2026,42(02):120-134. DOI： 10.11959/j.issn.1000-0801.2026016.

摘要

针对复杂背景下的人脸图像修复任务中普遍存在的细粒度纹理合成不足、结构修复断层和语义失谐的现象，提出了基于动态门控机制与自注意力模块融合增强的人脸图像修复网络。新算法通过构建多级膨胀卷积组捕获局部细节与长程上下文信息，并引入双重创新机制：（1）深度动态门控机制采用多层卷积与批归一化实现空间自适应的特征选择，取代传统残差连接的固定融合方式，显著提升了特征表达的灵活性和精准度；（2）自注意力机制显式建模全局像素依赖关系，有效解决了大范围缺损修复中的结构连贯性和细粒度纹理合成难题。实验结果表明，相对于较优对比算法SCAT，新算法在FFHQ、CelebA-HQ和LFW人脸数据集上的PSNR和SSIM指标平均提升了0.382 dB和0.004 1，FID平均改善了7.81%，尤其是在大面积遮挡（>50%）场景下，FID平均下降了2.153 4，显著提升了复杂背景下人脸图像修复质量，在生成逼真纹理、结构一致性方面有突出的修复优势。

Abstract

Aiming at the phenomena of insufficient fine-grained texture synthesis

structural repair faults

and semantic detuning

which are commonly found in face image restoration tasks in complex contexts

a face image restoration network based on the fusion enhancement of a dynamic gating mechanism with a self-attention module was proposed. The algorithm captured local details and long-range contextual information by constructing a multilevel dilated convolutional group

and introduced a dual innovative mechanism: (1) the deep dynamic gating mechanism adopted multilayer convolution with batch normalization to achieve spatially adaptive feature selection

replacing the fixed fusion of the traditional residual connection

which significantly enhanced the flexibility and accuracy of feature expression; (2) the self-attention mechanism explicitly modeled global pixel dependencies

which effectively solved the difficulties of structural coherence and fine-grained texture synthesis in large-scale defect repair. Experiments show that

compared with the better comparison algorithm SCAT

this new method improves PSNR and SSIM metrics by an average of 0.382 dB and 0.004 1

and improves FID by an average of 7.81% on three face datasets

namely

FFHQ

CelebA-HQ

and LFW

especially in the scene of large-area occlusion (>50%)

the FID decreased by an average of 2.153 4

significantly improving the accuracy of face images in complex backgrounds. It improves the quality of face image restoration under complex backgrounds

especially in generating realistic textures and structural consistency

showing outstanding advantages.

关键词

Keywords

references

Bertalmio M , Sapiro G , Caselles V , et al . Image inpainting [C ] // Proceedings of the 27th Annual Conference on Computer Graphics and Interactive Techniques-SIGGRAPH '00 . New York : ACM Press , 2000 : 417 - 424 .

Criminisi A , PéRez P , Toyama K . Region filling and object removal by exemplar-based image inpainting [J ] . IEEE Transactions on Image Processing , 2004 , 13 ( 9 ): 1200 - 1212 .

Brooks T , Holynski A , Efros A A . InstructPix2Pix: learning to follow image editing instructions [C ] // Proceedings of the 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . Piscataway : IEEE Press , 2023 : 18392 - 18402 .

Cao A Q , Dai A , De Charette R . Pasco: urban 3D panoptic scene completion with uncertainty awareness [C ] // Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . Piscataway : IEEE Press , 2024 : 14554 - 14564 .

Criminisi A , Perez P , Toyama K . Object removal by exemplar-based inpainting [C ] // Proceedings of the 2003 IEEE Computer Society Conference on Computer Vision and Pattern Recognition . Piscataway : IEEE Press , 2003 : II .

Wang X L , Girshick R , Gupta A , et al . Non-local neural networks [C ] // Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition . Piscataway : IEEE Press , 2018 : 7794 - 7803 .

Yu J H , Lin Z , Yang J M , et al . Generative image inpainting with contextual attention [C ] // Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition . Piscataway : IEEE Press , 2018 : 5505 - 5514 .

Yu J H , Lin Z , Yang J M , et al . Free-form image inpainting with gated convolution [C ] // Proceedings of the 2019 IEEE/CVF International Conference on Computer Vision (ICCV) . Piscataway : IEEE Press , 2019 : 4470 - 4479 .

Cai W W , Wei Z G . PiiGAN: generative adversarial networks for pluralistic image inpainting [J ] . IEEE Access , 2020 , 8 : 48451 - 48463 .

Wang N , Zhang Y P , Zhang L F . Dynamic selection network for image inpainting [J ] . IEEE Transactions on Image Processing , 2021 , 30 : 1784 - 1798 .

Yu Y C , Zhan F N , Lu S J , et al . WaveFill: a wavelet-based generation network for image inpainting [C ] // Proceedings of the 2021 IEEE/CVF International Conference on Computer Vision (ICCV) . Piscataway : IEEE Press , 2021 : 14094 - 14103 .

Zeng Y H , Fu J L , Chao H Y , et al . Aggregated contextual transformations for high-resolution image inpainting [J ] . IEEE Transactions on Visualization and Computer Graphics , 2023 , 29 ( 7 ): 3266 - 3280 .

Feng X , Pei W J , Li F J , et al . Generative memory-guided semantic reasoning model for image inpainting [J ] . IEEE Transactions on Circuits and Systems for Video Technology , 2022 , 32 ( 11 ): 7432 - 7447 .

Zuo Z W , Zhao L , Li A L , et al . Generative image inpainting with segmentation confusion adversarial training and contrastive learning [J ] . Proceedings of the AAAI Conference on Artificial Intelligence , 2023 , 37 ( 3 ): 3888 - 3896 .

Zhang H , Goodfellow I , Metaxas D , et al . Self-attention generative adversarial networks [C ] // Proceedings of International Conference on Machine Learning . Maastricht : PMLR , 2019 : 7354 - 7363 .

Karras T , Laine S , Aila T M . A style-based generator architecture for generative adversarial networks [C ] // Proceedings of the 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . Piscataway : IEEE Press , 2019 : 4396 - 4405 .

Huang G B , Mattar M , Berg T , et al . Labeled faces in the wild: a database forstudying face recognition in unconstrained environments [J ] . Computer Science , 2008 : 1 - 11 .

Karras T , Aila T , Laine S , et al . Progressive growing of GANs for improved quality, stability, and variation [PP ] . arXiv ( 2018-02-26 )[ 2025-03-11 ] arXiv: arXiv. 1710.10196.

Liu G L , Reda F A , Shih K J , et al . Image inpainting for irregular holes using partial convolutions [C ] // Computer Vision-ECCV 2018 . Cham : Springer , 2018 : 89 - 105 .

Guo X F , Yang H Y , Huang D . Image inpainting via conditional texture and structure dual generation [C ] // Proceedings of the 2021 IEEE/CVF International Conference on Computer Vision (ICCV) . Piscataway : IEEE Press , 2021 : 14114 - 14123 .

Xia X B , Yang W H , Ren J , et al . Pluralistic image completion with Gaussian mixture models [C ] // Proceedings of the Neural Information Processing Systems (NeurlPS 2022) . Piscataway : IEEE Press , 2015 : 1 - 14 .

Li X G , Guo Q , Lin D , et al . MISF: multi-level interactive Siamese filtering for high-fidelity image inpainting [C ] // Proceedings of the 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . Piscataway : IEEE Press , 2022 : 1859 - 1868 .

Wang Z , Li K , Peng J . Dynamic context-driven progressive image inpainting with auxiliary generative units [J ] . The Visual Computer , 2024 , 40 ( 5 ): 3457 - 3472 .

浏览量

202

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

基于深度学习的多样化复杂网络影响力节点识别

基于自注意力机制的大规模MIMO信道状态信息特征向量反馈方法