Spatial domain steganographic method detection using kernel extreme learning machine
Shaveta Chutani, Anjali Goyal · 2020
Recovering the hidden secret message in any stego image requires knowledge of the embedding algorithm. The steganography literature is abounding with different embedding paradigms, and multiplicity of such slightly different algorithms makes this task even more difficult. This paper proposes to determine the type of a message embedding steganographic method as a multi-class classification problem. We propose using kernel extreme learning machine (KELM) as a machine learning tool to learn the varied impacts caused by different steganographic methods on the underlying statistical interdependence of cover image pixels. To evaluate the performance of KELM, experiments are conducted on stego images created from BOSSBase v1.01 image dataset using two classical spatial domain Least Significant Bit embedding methods and two modern and adaptive spatial domain steganographic methods. The results indicate that KELM has significant potential in the multi-class classification of these embedding algorithms.