A Robust Bio-Signal Steganography With Lost-Data Recovery Architecture Using Deep Learning

Abstract

Information security is a major worry in today’s world, especially when e-health monitoring and cloud storage have become more widespread. Personal information of patients may be manipulated with or leaked as a result of network fault or an unethical service provider, causing significant disruption to clinical processes and compromising to patients’ privacy. In this paper, deep learning is used to present a resilient, imperceptible, and trustworthy biological signal steganography technique. The single channel signals and secret information are first turned into a chain of blocks, after which data insertion and extraction are carried out in the Hermite domain using a new technique, if no data is lost. If data loss is discovered in the cover signal during decryption, a particle swarm optimization-based estimate approach was presented for retrieving the lost secret information as well as the lost block feature. In addition, a supervised long-short term memory recurrent neural network (Su-LSTM) was devised, which effectively anticipated the lost signal. The technique was tested on a variety of ECG, PPG, and EEG signals, and the distortion error after encryption was found to be quite low. Following data extraction, the error was further decreased (percent root mean squared difference less than 0.01%), resulting in a partially reversible procedure. This proposed architecture allows for independent and multiple steganography in a single signal, as well as the prediction of missing blocks and lost data, with high storage capacity.