An Advanced Spiking Neuron Model And Its Application In Image Encryption

According to the basic computational units, artificial neural network can be distinguished as three generations. As the third artificial neural network, the networks of spiking neuron are computationally more powerful and more closed to biological neurons networks than the former two generations. Thus, it is necessary to design a spiking neuron model which is more closed to biological neuron and can be implemented in hardware. Combined with the design ideas of the Integrate-and-Fire model, an advanced spiking neuron model, that is the advanced spiking neuron model(ASNC), and an advanced pulse-coupled spiking neuron model, that is the advanced pulse-coupled spiking neuron model(APSNC), are proposed in this paper.Meanwhile, with the development of the digital age, the security issues of the digital information have become more prominent. Especially about digital image, its two-dimensional nature, high correlation, large amount and high redundancy of the information make some conventional encryption schemes be powerless. So it is necessary to improve the conventional encryption schemes or put forward a new encryption technology, to deal with the security of image information. Under such background, the chaos-based image encryption scheme is widely adopted because of its ability to effectively combine the sensitivity, ergodicity and complexity of chaos.After further research on the APSNC model, the suitability of its chaotic behavior for image encryption is found. Furthermore, a new encryption algorithm based on chaos and bit-level concentric rotation is proposed.The main works are as follows:1) Considering the memory of biological neurons, an advanced spiking neuron model with memory, which has a short memory for each reset threshold, is designed.Furthermore, the APSNC is realized in this paper, by cross-switching of the two ASNCs. In the APSNC, a short cross memory behavior is showed in the process of mutual control between the two ASNCs. In addition, the inputs of multiple periodic signals makes the neuron model be more general, but it also increases the difficulty ofcomputing the firing phase. In this paper, the method by extending the definition of least common multiple to get the common period of different signals, and further to obtain the ignition phase, is put forward. This method can be applied to compute the common period of the neuron model which has several input signals with different periods. Then the dynamic of the ASNC and APSNC are analyzed, and the parameters settings are divided into three regions with different dynamics. It is useful to obtain the desired dynamic behavior quickly. And the results of statistical analysis show that the chaotic sequences generated by the ASNC and APSNC have good pseudo randomness. It supports the feasibility of applying the ASNC and APSNC in image encryption. Furthermore, the Lena image is encrypted with the ASNC and APSNC.Besides, the hardware circuits of the ASNC and APSNC are built, and the corresponding tests are finished.2) A new encryption algorithm based on chaos and bit-level concentric rotation is proposed, using the Confusion-Diffusion structure that regards bits as basic computational units. The encryption algorithm first divides every 8 adjacent pixels into 8?8 bits block, and then the 4 regions from the inner to the outer of the block are rotated according to the instructions from chaotic sequence. This operation not only confuses the pixels values effectively, but also has part of the diffusion ability, and can achieve a good balance of gray histogram. So it shares the workload of Diffusion phase. By XOR and Mod operation, the proposed encryption algorithm can strengthen some link between the pixels of original image and encrypted image, which can make the slight change of a pixel of the original image spread to the whole encrypted image.The algorithm can also used to encrypt the RGB image through appropriately extending. In addition, it is found that a chaotic behavior, which has the continuous range of parameter setting and the uniform data distribution of iterations, after the study of the APSNC model. Therefore, the chaotic sequences used in encryption are generated by the APSNC. In the experiment, the grayscale, binary and RGB images are encrypted by the proposed algorithm. The encrypted results perform excellent in the visual discrimination, histogram, key space, key sensitivity, information entropy,correlation, and can withstand statistical attack, differential attack and blind decryption effectively.