Browsing by Author "Progonov, Dmytro"

Now showing 1 - 2 of 2

Detection of stego images with adaptively embedded data by component analysis methods
(Lviv Politechnic Publishing House, 2021) Progonov, Dmytro; Igor Sikorsky Kyiv Polytechnic Institute
Ensuring the effective protection of personal and corporate sensitive data is topical task today. The special interest is taken at sensitive data leakage prevention during files transmission in communication systems. In most cases, these leakages are conducted by usage of advance adaptive steganographic methods. These methods are aimed at minimizing distortions of cover files, such as digital images, during data hiding that negatively impact on detection accuracy of formed stego images. For overcoming this shortcoming, it was proposed to pre-process (calibrate) analyzed images for increasing stego-to-cover ratio. The modern paradigm of image calibration is based on usage of enormous set of high-pass filters. However, selection of filter(s) that maximizes the probability of stego images detection is non-trivial task, especially in case of limited a prior knowledge about embedding methods. For solving this task, we proposed to use component analysis methods for image calibration, namely principal components analysis. Results of comparative analysis of novel maxSRMd2 cover rich model and proposed solution showed that principal component analysis allows increasing detection accuracy up to 1.5 % even in the most difficult cases (low cover image payload and absence of cover-stego images pairs in training set).
Performance analysis of stego image calibration with the usage of denoising autoencoders
(Видавництво Львівської політехніки, 2022-06-06) Progonov, Dmytro; Igor Sikorsky Kyiv Polytechnic Institute
Methods for early detection of sensitive information leakage by data transmission in open (public) communication systems have been of special interest. Reliable detection of modified (stego) cover files, like digital images, requires usage of computation-intensive methods of statistical steganalysis, namely covering rich models and deep convolutional neural networks. Necessity of finetuning parameters of such methods to minimize detection accuracy for each embedding methods has made fast retrain of stegdetectors in real cases impossible. Therefore, development of low-complexity methods for detection of weak alterations of cover image parameters under limited prior information about used embedding methods has been required. For solving this task, we have proposed to use special architectures of artificial neural networks, such as denoising autoencoder. Ability of such networks to estimate parameters of original (cover) image from the noisy ones under limited prior information about introduced alterations has made them an attractive alternative to state-of-the-art solutions. The results of performance evaluation for shallow denoising autoencoders showed increasing of detection accuracy (up to 0.1 for Matthews correlation coefficient) in comparison with the state-of-the-art stegdetectors by preserving low-computation complexity of network retraining.