A new chaotic signal based on deep learning and its application in image encryption

To improve the security of image encryption in singular chaotic systems, an encryption algorithm based on deep-learning is proposed in this paper. To begin with, the chaos sequence is generated by using a hyperchaotic Lorenz system, prior to creating new chaotic signals based on chaotic characteristics obtained from he simulations of the powerful complex network structure of long-short term memory artificial neural network (LSTM-ANN). Then, dynamic characteristics of the new signals are analyzed with the largest Lyapunov exponent, 0-1 test, power spectral analysis, phase diagrams and NIST test. In the end, the new signals are applied to image encryption, the results of which verify the expected increased difficulty in attacking the encrypted system. This is attributable to the differences of the new signals generated using the proposed method from the original chaotic signals, as well as arises from the high complexity and nonlinearity of the system. Considering its ability to withstand common encryption attacks, it is hence reasonable to conclude that the proposed method exhibits higher safety and security than other traditional methods.