JPEG Steganalysis Based on Local Dimension Estimation
Xiaomei Quan · 2021
In this paper, a novel JPEG steganalysis scheme using local dimension estimation is proposed. Most steganalysis methods are based on the Cachin's statistical law. However, the high dimensionality of the feature space as well as the lack of the universal image statistical models limits the current steganalysis. The proposed method regards the difference between the host and stego images as a manifold and estimates its local dimension. By comparing the local dimensions of various data hiding methods, the proposed scheme can decide both the data-hiding method and the embedding rate. Experimental results demonstrate the effectiveness of the algorithm.