A New Technique of Cipher Type Identification Using Convolutional Neural Networks

This paper focuses on modern cipher types classification mechanism using convolutional neural networks (CNNs). In case of ciphertext-only-attack, it is momentous to a cryptanalyst to recognize the cipher type first for further cryptanalysis work, usually when the messages are transmitted from an unknown source. In this research paper, two modern ciphers named as AES-128 and RC-4 have been considered for the classification problem. At first, a corpus is created. Messages from the corpus are encrypted using these two encryption algorithms, and the ciphertexts are recorded into punched tape. We have taken the images of punched tapes and put those images into the dataset. These images are provided to different pretrained CNN architecture models to train the network models. For the training purpose, we have used transfer learning method. The results obtained from two different CNN architectures such as ResNet50V2 and MobileNetV2 have been compared and plotted. The primary success of this research is that a very much complex problem is solved in a much simpler way. The experimental results prove the validity of our proposed work.