A Lightweight Embedding Probability Estimation Algorithm Based on LBP for Adaptive Steganalysis

Adaptive steganography is the most advanced steganography currently, an important method to detect it is to integrate the embedding probability into feature extraction of adaptive steganalysis. Unfortunately, most of the existing methods directly use the true embedding probability maps, which are generated by prior knowledge: the specific steganographic strategies and embedding payloads. However, these cannot be known in advance for steganalysis tasks in the real world. To overcome this difficulty, we propose an embedding probability estimation algorithm based on the local binary pattern (LBP) for adaptive steganalysis. The algorithm we proposed has the advantage of not relying on prior knowledge. Meanwhile, for the first time, LBP operator is introduced into embedding probability estimation. As a non-machine learning method, it has a lighter-weight architecture because it does not need large-scale data sets for training. Experimental results show that the algorithm can better reduce the impact of embedding payloads mismatch than the existing methods, especially when the embedding payload is small.