Performance of Feature Extraction Techniques using CC-PEV and Subtractive Pixel Adjacency Matrix in Machine Learning-Based Steganalysis
Fajrul Malik Aminullah Napitupulu, Elviawaty Muisa Zamzami, Suyanto Suyanto · 2025
Steganalysis involves detecting hidden messages embedded within cover objects, such as images, audio, or video. This paper presents a universal steganalysis method aimed at countering the most widely used spatial-domain steganography techniques. The proposed model leverages support vector machines (SVM) to classify stego images based on extracted features. To evaluate the model’s effectiveness, we applied various feature extraction techniques and tested the steganalysis approach on a dataset of stego images. Our results show that the proposed method achieves robust performance in detecting stego images across different scenarios. Specifically, the model with the linear kernel and SPAM features yielded the highest detection accuracy, with an average of 75.28%. In contrast, the combination of the Gaussian kernel and CC-PEV features was less effective, achieving a lower detection accuracy of only 66.95%. These findings highlight the importance of selecting the right kernel and feature extraction method for optimal steganalysis performance. The study also suggests that while the linear kernel with SPAM features provides stable and reliable results, more complex methods like the Gaussian kernel require further refinement for better accuracy. Overall, this paper offers valuable insights into the effectiveness of different steganalysis techniques for combating modern steganography methods.