An Efficient System for Secret Information Sharing Through Machine Learning Based Key Generation and Steganography

Currently several cryptography methods are currently being developed, including chaos encryption, advanced encryption standard, two-fish, and others. Two major issues plague these cryptography algorithms: computational capabilities and sluggish learning. In order to address these concerns, this research article proposes a new cryptographic scheme. Steganography is a popular and dynamic technique for hiding important information or data within an image, video, or audio so that it cannot be accessed by unauthorized people. In this technique, it is planned to include number of methodologies to propose a new technique for gray and color images to produce better results with respect to efficiency and payload capacity. In this proposed technique first we have to obtain codeword with sensitive secret data with the help of its checksum, then the produced codeword is compressed with the suitable compression algorithm before encrypting, then it is added to the header and then inserted into the original image. To embed each byte of data combination of different LSB and MSB of the selected pixels is identified. The use of the Artificial Neural Network (ANN) algorithm in a cryptosystem has been found to improve cryptographic performance in terms of security and attack resistance. For one hidden layer NN, we describe a sub-key generation technique based on an Extreme Learning Machine (ELM) for producing a good cryptosystem. To initialize the input-hidden layer weights and data in each cycle, the initial key was built using the ANN topology, activation function, and seeds for the Pseudo Random Number Generator (PRNG).The output layer weights are used to construct the sub-key in each round. The proposed method is evaluated and compared to existing algorithms on a variety of images of varying sizes. The proposed algorithm produces better PSNR (Peak Signal to Noise Ratio), MSE (Mean Square Error), Average difference(AD), Maximum difference(MD), Normalized absolute error(NAE), Cross – correlation(CC) values, as well as better PSNR and MSE values, than the sequential algorithms for different embedding rates of 10%, 30%, and 50%.