Abstract:
Over the last twenty years, steganography has attracted in the consideration of numerous
information security specialists who analyzes and identifies possible threats
of information security. In contrast to encryption, which aims to secure the message,
steganography requires that a public message or image hides the fact of containing
any confidential data. Usually in standard methods of hiding information, the hidden
message is encoded in the least significant bit or bits. In the last decade, different
efforts have been made utilizing machine learning approaches to stow away data
without compromising the astuteness of the layer that transports data. Images are
frequently utilized as portion of communication since they are simpler to understand
than just a lot of routine text. Users can communicate over compromised networks,
in which case information security is essential. Subsequently, an effective method of
hidding information without alerting potential attacker about hidden information is
essential to guarantee the security of the data transmitted. In my thesis, I focus on
the research in image data steganography, which hides a confidential (secret) image in
another (cover) image. A deep convolutional neural network with a pixel permutation
is used as a hiding network, and various training schemes are evaluated to estimate
impact of the secret and cover image similarity on the efficiency of retrieval. My
main goal is to provide a method of image encryption that uses machine learning and
encryption technology.