High-Capacity Reversible Data Hiding using Deep Learning

In today's time, it has become imperative to lay more focus on the security problems such as privacy, copyright protection, and tamper detection which become relevant with the transfer of digital data. A strategy that has recently gained popularity is reversible data hiding (RDH), which helps the recipient on the other end to retrieve confidential information that the sender had concealed in digital data such as photographs or videos. This is done with the help of image encryption and Deep Neural Network (DNN). Through this study we have tried to dive deeper into the domain of reversible data hiding to try and gauge its effect when used with deep learning techniques on the decryption end. Images containing secret data were encrypted and then a deep learning-based model was utilized at the receiver end to classify which of the images had been decrypted accurately along with extraction of the hidden data bits. We have achieved higher embedding capacity with DNN when compared to the existing Support Vector Machine (SVM) scheme.