Steganography Detection using Convolutional Neural Networks: A Deep Learning Approach

• Published in: 2025 5th International Conference on Pervasive Computing and Social Networking (ICPCSN) pp. 981 - 985
• Main Authors: G, Saravana Gokul, Kumar K, Deepak, S, Senthil Pandi, P, Kumar, S, Monika, U, Neha M
• Format: Conference Proceeding
• Language: English
• Published: IEEE 14.05.2025
• Subjects: Accuracy; AlexNet; CNN; Computer architecture; Convolutional neural networks; Data models; Deep learning; Digital images; LeNet-5; Measurement; Pervasive computing; Social networking (online); Steganalysis; Steganography; Steganography Detection
• DOI: 10.1109/ICPCSN65854.2025.11035753

Steganography is a method of concealing any message or information within digital images, making it challenging to detect hidden messages. This study focuses on developing a deep learning-based approach for the detection of steganography with the help of Convolutional Neural Networks (CNNs). We implement and compare the performance of LeNet-5 and AlexNet architectures in classifying images as either clean or stego. The dataset comprises images embedded using the Least Significant Bit (LSB) technique. The models are then trained with augmented image data and then evaluated using various accuracy metrics. Experimental results indicate that AlexNet, with its deeper architecture, achieves higher accuracy than LeNet-5 in identifying stego-images. The findings demonstrate the potency of deep learning in automated steganalysis, highlighting the potential for CNNs in cybersecurity applications.