Toward the Confidential Data Location in Spatial Domain Images via a Genetic-based Pooling in a Convolutional Neural Network

Steganalysis, detecting hidden information within digital images, is crucial for securing data transmission. While current information security research focuses on non-adaptive steganography, there is a gap in addressing the challenge of locating payloads embedded through adaptive steganographic algorithms. This article introduces a novel steganalysis approach for the spatial domain. It identifies modification maps between two stego images and utilizes them as inputs to a convolutional neural network for pixel classification. Experimental assessments against adaptive steganographic algorithms WOW and S-UNIWARD consistently demonstrate superior performance, affirming the efficacy of the proposed strategy over existing methods. This methodology not only upholds network policies but also addresses the intricacies of adaptive steganography, enhancing overall security in digital image transmission.