摘要
基于最佳均方逼近,采用Hermite正交多项式做为神经网络隐层的激励函数,引入一种新型的Hermite神经网络模型.通过神经网络权值和混沌初值产生性能接近于理论值的混沌序列,从中提取与明文等长的序列进行排序,将排序结果对明文置换后即可得密文.加密与解密信息完全隐藏于神经网络产生的混沌序列中,与混沌初值无显式关系,且只需改变混沌初值,便可实现"一次一密"异步加密,其安全性取决于混沌序列的复杂性和无法预测性.理论分析和加密实例表明,该加密算法简单易行,克服了混沌同步加密的诸多缺陷,具有良好的安全性.
This paper introduces a new Hermite neural network model, in which orthogonal Hermite polynomials were employed as the activation functions of hidden layers of feed-forward neural networks by using best square approximation theory. By varying the chaotic initial value and regarding it as the input of the networks, new chaotic series were generated, which were close to the theoretical values. From the new chaotic series, a sub-sequence with the same length as the plain-text was extracted and sorted. Then, by replacing the sorted results with the plain-text, cipher-text was produced. In the encryption system, the information needed in encryption and decryption was hidden in the chaotic series and has no obvious relationship with the initial chaotic value. Security depends completely on the complexity and unpredictability of the chaotic sequences. By varying the initial chaotic value, we can implement asynchronous one-time pad cipher encryption. Theoretical analysis and encryption tests proved that our algorithm is useful, simple and highly secure, with many advantages that a synchronous system can never achieve.
出处
《智能系统学报》
2009年第5期458-462,共5页
CAAI Transactions on Intelligent Systems
基金
国家自然科学基金资助项目(60775050)