Detect-TPE: A New Framework for Ideal Thumbnail-Preserving Encryption via Face Detection
Dong Xie, Yibo Zhang, Zebang Hu, Taochun Wang, Fulong Chen, Peng Hu · IEEE Internet of Things Journal · 2024
Thumbnail-preserving encryption (TPE) is one of the prominent cryptographic primitives to balance the usability and privacy for cloud images. Ideal TPE requires that the thumbnail of the plaintext image is exactly the same as that of the ciphertext. Although there are some methods to improve the efficiency of ideal TPE, the results are not satisfactory due to the existing methods being within the time-consuming rank-encipher framework. Combined with face detection algorithms, we introduce a new framework, called Detect-TPE, for constructing efficient ideal TPE schemes in this article. The framework uses several classic detection algorithms to find the region of privacy and then encrypts the detected regions by ideal TPE. We use many different face detection algorithms (e.g., BlazeFace and YOLOv5-Face), and found that the main factors affecting the execution time of Detect-TPE are the size of block, the accuracy, and the efficiency of the used face detection algorithm. The experimental results show that if we use deep neural network (DNN) for face detection and the size of block is 64, the encryption time is reduced by 55.84% and the space occupied by the ciphertext image is reduced by 35.27% compared with existing TPE schemes. Additionally, the proposed Detect-TPE framework can resist facial detection attacks if the size of the block is greater than 32, and the probability of success in resisting this attack exceeds 99.22%.