647 Distinguishing Lightweight Block Ciphers in Encrypted Images Girish Mishra #,@,* , S.K. Pal # , S.V.S.S.N.V.G. Krishna Murthy @ , Kanishk Vats $ , and Rakshak Raina ! # DRDO-Scientifc Analysis Group, Delhi - 110 054, India @ DRDO-Defence Institute of Advanced Technology, Pune - 411 025, India $ Delhi Technological University, Delhi - 110 042, India ! Bennett University, Greater Noida - 201310, India * E-mail: gmishratech28@gmail.com ABSTRACT Modern day lightweight block ciphers provide powerful encryption methods for securing IoT communication data. Tiny digital devices exchange private data which the individual users might not be willing to get disclosed. On the other hand, the adversaries try their level best to capture this private data. The frst step towards this is to identify the encryption scheme. This work is an efort to construct a distinguisher to identify the cipher used in encrypting the trafc data. We try to establish a deep learning based method to identify the encryption scheme used from a set of three lightweight block ciphers viz. LBlock, PRESENT and SPECK. We make use of images from MNIST and fashion MNIST data sets for establishing the cryptographic distinguisher. Our results show that the overall classifcation accuracy depends frstly on the type of key used in encryption and secondly on how frequently the pixel values change in original input image. Keywords: Deep learning; Cryptography; Cryptanalysis; Lightweight block ciphers; MNIST; Fashion-MNIST Defence Science Journal, Vol. 71, No. 5, September 2021, pp. 647-655, DOI : 10.14429/dsj.71.16843 © 2021, DESIDOC 1. INTRODUCTION In an era of IoT, lightweight block ciphers provide a powerful way of encrypting the user data to ensure much- needed privacy. Billions of gadgets might need their own encryption schemes and the adversary on the other hand will need to identify the used scheme. This anticipatory fact brings the need to construct a distinguisher from the adversary’s perspective. The distinguisher for distinguishing the ciphers will also help the designer’s point of view as it may help to assess the cryptographic strength of the ciphers. The researchers have generally tried to develop two types of distinguishers; one which distinguishes between a cipher and random data. The other one predicts the class of cipher data. The development of the frst type of distinguisher is based on the fact that an adversary should not be able to ascertain whether Oracle is sending the data through an encryption scheme or a random source. The second type of distinguisher tries to identify the encryption scheme used during the communication. Rivest 1 pioneered in exploring the possibilities of the connection between cryptography and machine learning. He emphasised over the fact that how one area can contribute ideas and techniques to the other. He further perceived machine learning and cryptanalysis as sister felds as both share similar concerns and notions. After generating theoretical interest with this landmark paper and the subsequent availability of the plentiful advanced computing resources and the better- established theories, the researchers explored ML applications in cryptography in more depth. With the rapid growth in the availability of afordable internet services to the vast population and simultaneously emerging multimedia technologies, the image and video data are being transmitted over the network in a substantial amount. Therefore, the protection of multimedia data is a vital requirement. The researchers have been coming with diferent encryption approaches for protecting the confdentiality of this data. For example, chaos-based image encryption 2 , chaotic maps 3 , cosine-transform-based chaotic system 4 , the combination of an elliptic curve with Hill Cipher 5 and AES 6 are some of these approaches. On the other side, there have been continuous eforts to mount cryptanalytic attacks on encrypted images 7,8 . Linus LAGERHJELM 9 , in his master thesis, used Convolutional Neural Networks (CNN) to perform classifcation tasks over encrypted MNIST image dataset 10 . He considered it a traditional image recognition task and showed the encrypted images to the network to predict the class label. In a 10-class (encrypted MNIST image dataset) problem, he achieved the success rate of 10% and 42% for images encrypted in CBC and ECB modes, respectively. The better results for ECB mode can be attributed to not having the desired randomness characteristic, as it is well established that ECB is the weakest mode of encryption. De mello 11 , et al. used machine learning techniques to identify encryption algorithms in a ciphertext-only setup. The plaintext corpora for Received : 22 February 2021, Revised : 05 April 2021 Accepted : 12 April 2021, Online published : 02 September 2021