A Deep Learning Based Image Steganalysis Using Gray Level Co-Occurrence Matrix

Abstract

Image steganalysis is the technique to identify steganography in images and if possible try to predict the quantity of hidden data. Targeted steganalysis need the knowledge of the steganographic algorithm used to embed the data, whereas blind steganalysis is independent of the embedding process. Its objective is to find patterns in the stego image that are generated due to steganographic process. This work proposed an elegant technique of blind steganalysis which takes input clean image from a benchmark dataset and find the co-occurrence matrix of grayscale image(GLCM) for four pixel pair direction and produces average GLCM. After that PCA and Haralick features are generated from average GLCM. Next, steganographic embedding is applied to the clean images with Steghide application with different payload. Each of this stego images are applied similar feature extraction process as clean images in the dataset. Now a deep neural network based model is trained on the prepared dataset with proper label. The proposed method gives 90.93% and 84.63% accuracy for LFW and BOSS dataset respectively and outperform many similar algorithms.