Abstract:
Information security is a top priority today. Steganography and cryptography are two such techniques that are used to ensure secure data transaction. By encrypting the data, cryptography changes it, making it impossible to understand and use. However, steganography completely hides the data's existence, thwarting all attacks. While steganography is a great tool for data protection, it has also been used to compromise security. While keeping data safe from unwanted access is important, obscuring its presence can help keep illegal information hidden. Although there are numerous documented steganography techniques, there are only a few ways to detect steganography or perform steganalysis. Additionally, there are many difficulties in creating classifiers for detection, including various steganographic techniques, the lack of the original file, and distribution disagreement across many domains. The focus of this research was to create a neural network capable of detecting steganography in compressed colour photos and introduce an efficient steganalysis method. This was done by researching the statistical effects of steganography on an image and choosing a set of hand-crafted attributes that could be used to detect the presence of steganographic embedding. A limited set of images consisting of unique steganography and regular images were used for training and validation. To develop the best classifier under the data and technology limitations, this method experimented with a wide range of features and training parameters. Finally, a series of enhancements that can be made by future researchers have been suggested.
