Résumé:
In an era where privacy has become increasingly important with the constant
informatisation of our day-to-day tasks, the quest to safeguard sensitive and personal
information had led to the invention of various methods. Throughout history, the
persistent need for secrecy and confidentiality has served as the driving force behind the
development of these methods, including encryption techniques, anonymization
protocols and secure communication systems. However, a paradoxical phenomenon has
emerged as these very tools, which were initially intended to protect privacy, are now
being exploited for the malicious purposes they were designed to guard against, one of
these techniques is steganography.
The misuse of steganography to conceal malware within innocent media files,
particularly images, has given rise to a significant cybersecurity concern known as
stegomalware or stegware for short. Threat actors have recognized the potential of
utilizing this technique to embed and distribute malicious payloads undetected.
Consequently, traditional measures and defences are rendered powerless in the face of
this sophisticated threat.
In this research, we aim to combine Deep Learning, Malware Analysis and
Steganalysis techniques in order to put in place a system capable of dissecting and
detecting stegware present specifically in PNG images. Our system comprises three main
components. Firstly, we implement various steganalysis deep learning models proposed
by researchers in the field, making the necessary adjustments and modifications to suit
our case of study. The purpose of this first model is to determine the presence of
steganography in images. Subsequently, we employ a module to extract hidden data from
images identified as steganographic. Lastly, a text-based classification model is utilized to
categorize the extracted data as either malicious or clean. The implementation details,
rigorous testing, and comprehensive results will be discussed and presented in this study.
Keywords: Steganography, Malware, PNG Images, Deep Learning, Malware Analysis, Steganalysis, Detection, Classification.
