Detection System for Grey and Colour Images Based on Extracting Features of Difference Image and Renormalized Histogram

The literature has introduced many steganalysis methods intended to com- bat specific steganography techniques and to detect particular image formats. This paper proposes a detection system based on extracting histogram features. The features are extracted by exploiting the histogram of difference image, which is usually a generalised Gaussian distribution centred at 0. The histogram of difference image and the renormal- ized histogram are created for clean and stego images, therefore using the peak value and renormalized histogram as features for classification. To obtain the difference between neighbouring pixels, the difference images are computed for four directions (vertical, hor- izontal, diagonal, and anti-diagonal). The renormalized histogram of the difference im- age is created a number of times (n) for the four directions. This work implements two commonly-used steganography methods: the Least Significant Bit (LSB) and F5 al- gorithm to create a large database of stego images for system evaluation. Colour and grey images with different formats are chosen for training and testing the system. These formats are lossless and lossy compressions, with all features extracted from each colour channel (RGB) separately. The size of hidden files plays an important role in terms of detection. Therefore, to improve the proposed systems detection capacity, different sizes of hidden files have been considered. The proposed detection system was trained and tested to distinguish stego images from clean ones using the Discriminant Analysis (DA) classification method and Multilayer Perceptron neural network (MLP). The ex- perimental results prove that the proposed system possesses reliable detection ability and accuracy. The chosen classification methods show dissimilar performance in terms of classifying grey and colour images. The system holds more generalisability than previous systems by covering different types of stego images, image formats and hidden file sizes. In addition, extensive experimental results show that the proposed steganalysis system outperforms some previous detection methods.