Impact on JPEG Image Steganalysis using Transfer Learning

Abstract

Image steganography is a convenient candidate for information hiding with the upsurge of using images on social media platforms and the internet. It is equally important that detecting these images through steganalysis so that the illegal activities could be monitored. In this work, approaches that aren’t widely explored in the image steganalysis domain were analyzed. The concept of Transfer learning using recent ImageNet models such as EfficientNet, MobileNetV3, DenseNet and MixNet and attempts on developing ensemble models using the above-pretrained architectures were experimented. Results and findings of these experiments are documented through this work. These findings on the unexplored areas could potentially help in building a universal image steganalysis tool for the domain.