Reduction of LSB detectors set with definite reliability

Abstract

The article focuses on decreasing the set of steganalytical methods that determine the payload value in the image spatial domain using quantitative detectors of Least Significant Bits (LSB) steganography. It is supposed that methods can trace the same image regularities, and hence their results can correlate. The work presents the results of the development and testing of the technique for reducing the set of steganalytical methods taking into account the accuracy and reliability to the diminution of the computational complexity of steganalytical expertise. The theoretical basis of the proposed solution is the approximation of regression of the first kind by linear regression of the second kind for multivariate random variables. To verify the results, the computational experiment was performed. The payloads were implemented in 10 % increments by automating the freeware steganographic programs CryptArkan and The Third Eye with AutoIt. The steganalytical methods, such as Weighted Stego, Sample Pairs, Triples analysis, Asymptotically Uniformly Most Powerful detection, Pair of Values, were used. The datasets were built in the MATLAB environment; the program was implemented in Python. For the experiment’s reproducibility, the datasets and program code are provided in Kaggle. Interval estimates of methods correlation are calculated based on experimental data for various payload values. The developed technique includes a mathematical model, an algorithm for implementing the model, and a computer program. The proposed technique can be applied in those tasks where accuracy and reliability are taken into account. One of the subject areas demanding such assessments is computer forensics dealing with expertise with probabilistic conclusions. These estimates allow the analyst to vary the number of methods depending on the available computing resources and the time frame of the research.