基于离散型Hopfield神经网络的图像加密算法

Image Encryption Algorithm Based on Discrete Hopfield Neural Network

为了提高加密算法的安全性能,该文提出了明文相关的通道关联的同时置乱扩散加密方案,算法中置乱扩散矩阵是与明文图像以及外部密钥相关的。由于明文图像中相邻像素之间的相关性很高,因此,图像加密方案一般采用置乱的方式对图像进行"洗牌"操作,再进行扩散操作改变像素值,即Fridrich提出的经典置乱-扩散结构。然而随着图像破解技术的进步,分开攻击技术对这种经典的加密方案造成了很大的影响。该文采用同时置乱扩散的方案提高加密算法的安全性,首先使用三阶神经网络产生密钥流对彩色图像的三通道进行同时置乱扩散操作,然后再对得到的三个通道矩阵进行二次扩散操作,使得加密算法更加安全。对比现有的图像加密方案,该加密算法能够避免分开攻击对加密安全性的影响,提高了算法的安全性。实验结果表明,该算法有良好的加密性能,可以抵抗各种攻击。

In order to improve the security performance of the encryption algorithm, we propose a simultaneous scrambling and diffusion encryption scheme that is plaintext related and channel associated. The scrambling diffusion matrix in the algorithm is related to plaintext image and external key. Since the high correlation between adjacent pixels in plaintext image, image encryption schemes generally use scrambling to permutate the image, and then diffusion operation can change the pixel value, which is the classical permutation-diffusion structure proposed by Fridrich. However, with the progress of image cracking technology, the separate attack technology has a great impact on this classic encryption scheme. We adopt the scheme of simultaneous permutation and diffusion to improve the security of encryption algorithm. First of all, we use the third-order Hopfield neural network to generate the key stream to permutate and diffuse simultaneously for the three channels of the color image, and then carry out the secondary diffusion operation on the three matrices, which makes the encryption algorithm more secure. Compared with the existing image encryption schemes, the proposed algorithm can avoid the impact of separate attacks on the encryption security and improve the security of the algorithm. Experimental results and security analyses verify that the proposed scheme can achieve ideal encryption result and resist various attacks.