ADAPTIVE ALGORITHM OF REPLACEMENT-BASED EMBEDDING OF DATA INTO COMPRESSED JPEG IMAGES

Abstract

Introduction: Digital steganography is an effective approach to ensuring the confidentiality of transferred and stored information. In order to provide stability before steganalysis when data are embedded into digital images, it is important to avoid the appearance of unmasking features caused by the embedding. Purpose: Developing an adaptive algorithm of steganographic embedding of data into compressed JPEG images based on a replacement operation, minimizing the distortions introduced to the informative features. Results: The paper discusses the importance of unmasking features in steganalysis and their application for adaptability of information concealment algorithms in digital objects. The main features in the spatial and frequency domains of digital images applied in modern steganographic embedding methods are specified. Informative features are selected, excluding linearly dependent features or features without any information about digital object distortion during the embedding. The resulting set has allowed us to increase the accuracy of general classification of images by 19%. On the base of the obtained set of informative features, a replacement-based adaptive modification has been developed for the algorithm of embedding data into compressed JPEG images. This modification minimizes the image container distortions during the embedding due to the use of а criterion function formulated in the article. The algorithm is adaptive because the concealment field is chosen based on the set of informative features which characterize a natural model of a digital image. Computing experiments allowed us to find the best parameter values in order to achieve good embedding capacity and the minimum distortions of the unmasking features. Experiments with the developed adaptive algorithm have demonstrated its increased stability before steganalysis and very good embedding capacity. Also, it has high values of the stego-image quality metrics, making the distortions less noticeable either for human eyes or for numerous steganalysis algorithms, because the values of unmasking features are distorted only slightly.