Feature Mining and Neuro-Fuzzy Inference System for Steganalysis of LSB Matching Steganography in Grayscale Images

Qingzhong Liu, Andrew H. Sung

In this paper, we present a scheme based on feature mining and neuro-fuzzy inference system for detecting LSB matching steganography in grayscale images, which is a very challenging problem in steganalysis. Four types of features are proposed, and a Dynamic Evolving Neural Fuzzy Inference System (DENFIS) based feature selection is proposed, as well as the use of Support Vector Machine Recursive Feature Elimination (SVM-RFE) to obtain better detection accuracy. In comparison with other well-known features, overall, our features perform the best. DENFIS outperforms some traditional learning classifiers. SVM-RFE and DENFIS based feature selection outperform statistical significance based feature selection such as t-test. Experimental results also indicate that it remains very challenging to steganalyze LSB matching steganography in grayscale images with high complexity.