Toward secret data location via fuzzy logic and convolutional neural network

Locating hidden data in digital images, otherwise called steganalysis, is a process of identifying the existence of secret messages within digital images. Steganalysis is used to manage digital data transmission by detecting the possible hidden information that can be used to violate the network policy; hence, it helps the development of policies and regulations aimed at strong protection from cyber threats to individuals’ and organizations’ data. The research works in the field of information security commonly focus on developing the locating approaches for non-adaptive steganography, which present a problem of less investigation of the complex challenge of locating the payload embedded with an adaptive steganographic algorithm. In this article, we propose a method to locate hidden data in a digital image in three stages: a) Identification of the modification maps between the carrier and final images. b) Using the modification maps as input to Mamdani fuzzy inference with four input membership functions: covariance map matrix, compass mean matrix, distance vector matrix, and pixel intensity matrix, and one output membership function, notably the fuzzy correlation maps. c) Feeding the fuzzy correlation maps to a convolutional neural network to identify the pixels with confidential data from the innocent pixels. By experimenting with our method against four steganographic algorithms, namely, HILL, HUGO-BD, WOW, and S-UNIWARD, the recall rates for the four algorithms initially increase in a similar range and improve with increasing payload capacity, which justifies the outperformance of the proposed strategy over the existing methods.