Visual Privacy Protection Method Based on YOLOv8 and ControlNet

To achieve a balance between visual privacy protection and visual information integrity, we propose a visual privacy protection method based on YOLOv8 and ControlNet. In the privacy recognition process, we realize the targeted selection of redrawing area by recognizing privacy features and effective...

Full description

Saved in:
Bibliographic Details
Published in2024 5th International Conference on Computer Vision, Image and Deep Learning (CVIDL) pp. 1023 - 1029
Main Authors Li, Kesong, Yang, Guanci
Format Conference Proceeding
LanguageEnglish
Published IEEE 19.04.2024
Subjects
Accuracy
Computational modeling
ControlNet
Data privacy
Diffusion models
Privacy
Stable diffusion
Target recognition
Visualization
YOLOv8
Online AccessGet full text

Cover

More Information
Summary:To achieve a balance between visual privacy protection and visual information integrity, we propose a visual privacy protection method based on YOLOv8 and ControlNet. In the privacy recognition process, we realize the targeted selection of redrawing area by recognizing privacy features and effectively segmenting the people and backgrounds that leak privacy. In the privacy preservation process, individuals undergo virtual dressing through the ControlNet-guided Stable Diffusion model, resulting in visually protected images. Empirical studies demonstrate promising results, with an average accuracy of 94.7 \% in privacy identification. Additionally, the average Structural Similarity Index (SSIM) value between privacy-protected images and original images stands at 0.843
DOI:10.1109/CVIDL62147.2024.10604228