An Efficient Machine Learning Based Image Encryption Scheme for Medical Image Security

Abstract

Medical images are very important in most of the application than any other images. In real time applications like telemedicine application, communication of medical image through open access needs protection as well as security at high level. Many imaging information has its own unique features which are so difficult to analyse and make decision to identify necessary techniques for protecting confidential image of unauthenticated access, Utmost all the existing encryption algorithms are mainly concentrating on textual data, but for multimedia data like images, it is not suitable. The main contribution done in this work is for ensuring increased security level over medical images regardless of presence of noises. In this algorithm, DNA subsequence operations combining with the use of improved Combined Linear Congruential Generator (C-LCG) were used for encryption of information. This paper discuss the idea of the improvement of safe and secure techniques using machine learning which is justified by the entropy value and correlation among adjacent pixels with performance parameters. The original image was scrambled using Combined Linear Congruential Generator with Bit rotation operation (BRO) and then image is transformed by effective encryption method using DNA subsequence operations. The proposed scheme discloses the correlation between pixel and entropy. Experimentation results showed that correlation among pixels is reduced while maximizing entropy. Number of Pixel Change Rate (NPCR) and Peak signal to noise (PSNR) ratio were also been analysed. In proposed algorithm, maximum NPCR values is achieved which shows the scattering of pixels in Encrypted image is high. PSNR shows a better encryption quality with lower the values.