In the field of information security, block cipher is widely used in the protection of messages, and its safety naturally attracts people’s attention. The identification of the cryptosystem is the premise of encrypted data analysis. It belongs to the category of attack analysis in cryptanalysis and has important theoretical significance and application value. This paper focuses on the extraction of ciphertext features and the construction of cryptosystem identification classifiers. The main contents and innovations of this paper are as follows. Firstly, inspired by language processing, we propose the feature extraction scheme based on ASCII statistics of ciphertexts which decrease the dimension of data preprocessing. Secondly, on the basis of previous work, we increase the types of block ciphers to eight, encrypt plaintext of the same sizes as experimental objects, and recognize the cryptosystem. Thirdly, we use two machine learning classifiers to perform classification experiments including random forest and SVM. The experimental results show that our scheme can not only improve the identification accuracy of 8 typical block cipher algorithms but also shorten the experimental time and reduce the computation load by greatly minimizing the dimension of the feature vector. And the various evaluation indicators obtained by the scheme have been greatly improved compared with the existing published literature.

In reality, many cryptographic analysis techniques are based on a specific cryptographic system or a large number of encrypted ciphertext. The identification and detection of cryptographic system is of great significance for evaluating the security of the algorithm and guiding the design and improvement of the algorithm. In this paper, we transcode each character in ciphertext into a decimal number, construct these numbers into one-dimensional arrays, and obtain the Euclidean distance between these one-dimensional arrays. Then we use these distances as features and input them into three machine learning classifiers: random forest, logistic regression and support vector machine to recognize cryptosystem and compare their recognition accuracy. The subjects include 8 common block ciphers (DES, 3DES, AES-128, AES-256, IDEA, SMS4, Blowfish, Camellia-128). The experimental results show that using the feature extraction scheme not only shortens the experimental time, reduces the computational cost, but also improves the recognition accuracy of eight typical block cipher algorithms. The classification accuracy of the ECB mode in the random forest classifier is 75%, which is higher than the existing published literature experimental results. The classification accuracy rate of CBC mode is higher than 13.5%, which is higher than the accuracy of random classification.

Backpropagation neural network algorithms are one of the most widely used algorithms in the current neural network algorithm. It uses the output error rate to estimate the error rate of the direct front layer of the output layer, so that we can get the error rate of each layer through the layer-by-layer backpropagation. The purpose of this paper is to simulate the decryption process of DES with backpropagation algorithm. By inputting a large number of plaintext and ciphertext pairs, a neural network simulator for the decryption of the target cipher is constructed, and the ciphertext given is decrypted. In this paper, how to modify the backpropagation neural network classifier and apply it to the process of building the regression analysis model is introduced in detail. The experimental results show that the final result of restoring plaintext of the neural network model built in this paper is ideal, and the fitting rate is higher than 90% compared with the true plaintext.

No abstract