Image scrambling adversarial autoencoder based on the asymmetric encryption

Autor: Yadong Jin, Zhenjie Bao, Ru Xue
Rok vydání: 2021
Předmět:
Zdroj: Multimedia Tools and Applications. 80:28265-28301
ISSN: 1573-7721
1380-7501
DOI: 10.1007/s11042-021-11043-3
Popis: For the purpose of information transmission security, image scrambling is to encrypt the image by changing the image pixel values and pixel positions. Based on the asymmetric encryption, we propose a model of Image Scrambling Adversarial Autoencoder. Firstly, we describe an encoder-decoder framework to imitate the procedure of image scramble, descramble and the key generation. Secondly, we employ the generator network of CycleGAN as the encoder and decoder structure of our method to transfer the secret image to totally meaningless image and reconstruct it. Thirdly, the parameters of the encoder and decoder can be regarded as the public key and private key. Then, the patchGANs discriminator is used to distinguish encoded images and evenly distributed noise by image blocks. Moreover, we combine the encode-then-decode loss function with the adversarial loss function by an adjustable parameter in order to make the model training results more stable. Experiments show that our method can accomplish automatic image scrambling in ten different scenes which include Africa people and villages, beach, buildings, buses, dinosaurs, elephants, flowers, horses, mountains and glaciers, food. Compared with 3D Arnold transformation and CycleGAN, scrambled pixels by our method are more evenly distributed intuitively. What is more, extensive experiments show that the proposed method can address security requirements partly and achieve a good encryption efficiency. In addition, contrast experimental results show that the combination of the encode-then-decode loss function and the adversarial loss function is essential to achieve the ideal results of image scrambling and restoration.
Databáze: OpenAIRE
Externí odkaz: