International Research Journal of Engineering and Technology (IRJET) Volume: 09 Issue: 06 | Jun 2022
www.irjet.net
e-ISSN: 2395-0056 p-ISSN: 2395-0072
DETECTING THE SECURITY LEVEL OF VARIOUS CRYPTOSYSTEMS USING MACHINE LEARNING MODELS B.Manjunath1, K.S.Phaneendra2, M.Gowrish3 1,2,3 Pg
Scholar Department Of computer of application Institute of Technology and Science , Andhra Pradesh, 517352 ---------------------------------------------------------------------***--------------------------------------------------------------------insufficient to preserve its privacy. For example, if a single Abstract - The security of digital data has become a huge 1,2,3Madanapalle
substitution box (S-box) is used to encrypt an image, the information in the substituted or enciphered image may still be visible. This means that a single S-box encryption is insufficient to properly conceal the source image. Despite the fact that the information being communicated is encrypted, unauthorised individuals can still view it due to the encryption algorithm's flaws, as seen in Figure 1. (b). Thus. To improve encryption security, it is also vital to utilise a strong encryption algorithm. The security level of the encryption algorithm used to encrypt the image has a significant impact on its robustness. The plain image will be entirely encrypted using a highly strong encryption method, allowing it to withstand attacks on its integrity, secrecy, and availability. Along with security, temporal complexity is a significant consideration when choosing an effective encryption method. Because different types of data have different security priorities, choosing a cryptosystem is dependent on the nature of the application to be encrypted. The Advanced Encryption Standard (AES) [7] is, for example, the most secure encryption algorithm available at the moment. However, because AES requires numerous rounds, which takes longer, it is not suited for applications that demand quick encryption. The original information must be encrypted, which takes additional time. Furthermore, the total number of pixels in the source image influences the temporal complexity. The higher the amount of pixels in the plain image, the longer it will take to encrypt it [8]. If the main requirement is simply to encrypt a plain image with high security, on the other hand, processing time may not be as important. Although strong encryption gives superior security, it is not always a property of quick encryption, which is sometimes preferred [9]. A statistical study such as entropy, correlation, energy, or homogeneity must be done on an encryption algorithm to determine its security level. Testing each encryption technique and calculating the statistics of its security characteristics can help with these duties. We can choose the best and strongest choice from those examined after completing such security studies on each encryption method one by one. However, this procedure frequently detracts from the completion of the assignment. Instead, we recommend that manual testing be replaced by a machine learning model that can quickly, easily, and accurately identify the strongest encryption technique. Based on conventional security factors of encryption algorithms, we have classified the security of encryption algorithms into three levels (strong, moderate,
concern as a result of recent advances in multimedia technology. Researchers tend to focus their efforts on changing existing protocols to overcome the flaws of present security mechanisms. Several proposed encryption algorithms, however, have been proven insecure during the last few decades, posing serious security risks to sensitive data. Using the most appropriate encryption technique to protect against such assaults is critical, but which algorithm is most suited in any given situation will depend on the type of data being secured. Testing potential cryptosystems one by one to identify the best alternative, on the other hand, can take a long time. . We present a security level identification approach for picture encryption algorithms that incorporates a support vector machine for a fast and accurate selection of relevant encryption algorithms (SVM). We also generate a dataset with conventional encryption security criteria like entropy, contrast, homogeneity, peak signal to noise ratio, mean square error, energy, and correlation in this research. These parameters are collected from various cypher pictures as features. The security level of dataset labels is categorised into three categories: high, acceptable, and weak. We used various studies (f1-score, recall, precision, and accuracy) to assess the performance of our suggested model, and the findings show that this SVMsupported system is effective.
Keywords: Support vector machine (SVM), security analysis, image encryption, cryptosystem.
1. INTRODUCTION Security has become a much-in-demand topic of research because to the exponential increase in transmissions of multimedia data across insecure channels (primarily the Internet). Many researchers have turned to inventing new encryption methods to safeguard data from eavesdroppers and unauthorised users [1]–[5]. Two elements are critical when encrypting digital images: diffusion and confusion (also known as scrambling). Claud Shannon proposed in [6] that a cryptosystem that includes confusion and diffusion methods can be regarded secure. The scrambling process on digital images can be done directly on pixels or on rows and columns, whereas diffusion affects the original pixel values. In other words, the substitution procedure substitutes each unique pixel value with the S-unique box's value. The transmission of data in an encrypted format, however, is
© 2022, IRJET
|
Impact Factor value: 7.529
|
ISO 9001:2008 Certified Journal
|
Page 1060