부가적인 데이터 삽입을 이용한 CNN기반의 스테그아날리시스

Abstract

Image steganography is an information hiding algorithm that embeds secret data into a digital image. Steganography has been used in a variety of fields requiring information security, such as medicine, business and defence, and steadily studied. Along with the study of steganography, steganalysis, which identifies whether steganographic algorithms are applied, has also been actively studied. Steganalysis algorithms are classifed into handcrafted feature-based and convolutional neural network(CNN)-based ones. A handcrafted feature-based steganalysis algorithm extracts statistical features from images and learns them using machine learning algorithms to generate classifiers. On the other hand, CNN-based algorithms incorporate the process of automatically extracting features from images and effectively learning them by using CNN which proved its performance in many classification challenges. CNN-based steganalysis. Required preprocessing to enhance minute changes from images, unlike conventional problems such as image recognition. Most CNN-based steganalysis studies have improved the performance through use of effective preprocessing filters or structural optimization. In this thesis, we propose a new approach that is based on additional data embedding to an input image and uses dual inputs, the input image and its stego image, instead of trying to changing the preprocessing filters and network structure.