ABSTRACT With the popularity of the Internet and intelligent devices, the privacy protection of image data has become an important problem that needs to be solved urgently. Traditional privacy protection methods such as anonymization have limitations in processing image data and cannot balance data availability and privacy. Given this, an improved generative adversarial network image privacy protection model that combines facial feature enhancement, differential privacy, and gradient noise addition is constructed, aiming to effectively protect sensitive information in image data. The performance test results showed that when the dataset size was 1400, the average encryption and decryption time of the model were 1.68 and 1.52 s, respectively. The privacy protection success rates in the Top‐1 and Top‐10 protection success rate tests were 95.13% and 92.07%. In the four portrait restoration tests, the similarity between the restored image and the original image was 88.67%, 96.34%, 98.76%, and 92.47%, which were the best among the comparison models. The experiment shows that the proposed portrait privacy protection model outperforms existing methods in terms of privacy protection strength and image generation quality, and has good comprehensive performance.
Duan Xue (Fri,) studied this question.