Machine Learning Applied to the Recognition of Cryptographic Algorithms Used for Multimedia Encryption

This paper presents a study of encrypted multimedia files in order to identify the encryption algorithm. Audio and video files were encoded with distinct cryptographic algorithms and then metadata were extracted from these cryptograms. The algorithm identification is obtained by using data mining techniques. Therefore, the procedure first stage performs the encryption of audio and video files using DES, Blowfish, RSA, and RC4 algorithms. Then, the encrypted files were submitted to the data mining algorithms: J48, FT, PART, Complement Naive Bayes, and Multilayer Perceptron classifiers. The resulting confusion matrices compiled into charts and it was possible to notice that the percentage of identification for each of the algorithms is greater than a probabilistic bid. There are several scenarios where algorithm identification reaches almost full recognition.