差分隐私结合GAN的图像数据保护方法仿真

Simulation of Image Data Protection Method for Differential Privacy Combined with GAN

图像数据的过度挖掘会对图像隐私性带来影响,使得数据被泄露或滥用的风险不断增加。为避免图像隐私数据泄露,提出差分隐私结合GAN的图像隐私数据保护算法研究。针对当前图像数据参杂的不确定性,通过对图像数据清洗,获取的图像重要隐私区域,有利于后续噪声的加入,便于实现图像隐私数据的保护;根据差分隐私与生成对抗网络(Generative Adversarial Networks, GAN)结合,构建图像数据差分隐私保护的GAN模型,依据拉普拉斯实现机制,对清洗后的图像数据扰动,达到隐私数据保护的目的;由于模型在迭代期间会对图像数据隐私性产生侵犯,所以进一步对图像隐私性侵犯程度度量,令上述算法在实际应用时有效兼顾图像隐藏数据的安全性及隐私性。实验结果表明,所提方法对图像的加密和解密时间短,且隐私性高。

In order to avoid privacy data leakage,this article put forward an algorithm for protecting GAN image privacy data based on differential privacy.Aiming at the uncertainty of the mixed image data,we cleaned the image data and thus obtain important privacy areas of the image,which was beneficial for the subsequent addition of noise and the protection of image privacy data.Moreover,differential privacy was combined with Generative Adversarial Networks(GAN)to construct a GAN model for differential privacy protection of image data.According to the Laplacian implementation mechanism,the cleaned image data was perturbed to achieve privacy data protection.Due to the invasion of image data privacy during the iteration of the model,the degree of image privacy infringement was fur-ther measured to enable the algorithm to effectively balance the security and privacy of hidden data in practical appli-cations.Experimental results show that the proposed method has short encryption and decryption time for images,and high privacy.