A Deep Learning-Based Steganalysis Model for Color Images Using Statistical Texture Features

The increasing reliance on the internet and social media platforms has raised significant concerns about data security, especially in digital images. Typically, the combination of large data capacity, widespread use, complexity, and perceptual masking properties makes images a versatile and effective medium for hiding secret data through steganography. While steganography has legitimate uses such as secure communication and digital watermarking, it can be exploited for illegal activities. In this research work, we propose a steganalysis deep learning-based neural networks model to detect the presence of secret data in colored images using statistical features correlations and deep learning classification. To increase the automated detectability of the presence of hidden data in images, statistical texture features were extracted for each single channel of an image from a benchmark dataset of RGB color images by analyzing the spatial relationships between pairs of pixels in the image to generate a feature set using Gray Level Co-Occurrence Matrix (GLCM). This feature set is then fed into a deep learning steganalysis classification model using artificial neural networks (ANNs) and convolutional neural networks (CNNs) for hidden data detection. We used a validation approach and selected the best parameter values to enhance the performance of our deep learning steganalysis model. To evaluate the performance of the proposed models we used a public dataset of 10,000 uncompressed images using detection accuracy metrics. After conducting multiple experiments, the best result (Accuracy = 92.15%) was achieved using our CNN model.