Date of Graduation

Spring 2022


Master of Natural and Applied Science in Computer Science


Computer Science

Committee Chair

Mohammed Belkhouche


During the past decade, many methods have been introduced to handle the image-based steganalysis problem. Traditional steganalysis methods are based on the two-step machine learning mechanism that consists of extracting and classifying phases. Most recent solutions are based on deep convolution neural networks (CNNs), which combine feature extraction and classification in one step. CNN-based steganalysis methods provide superior performance. These CNNs are designed to improve the detection rate by using a set of predefined filters for the pre-processing phase. In this thesis, I propose a CNN model that consists of two convolution layers for pre-processing and features extraction, and four fully connected layers for classification. The pre-processing layer relies on some of the well-known and efficient filters that were used in previous studies in addition to the various instances of the 2D Gabor filter based on selected parameters. I conducted experiments using grayscale cover images from a popular and publicly available BOSSbase_1.01 database with a consideration for two different image sizes. The results showed that the proposed CNN model outperforms many of the state-of-the-art studies in two out of three well-known adaptative spatial domain steganography algorithms (S-UNIWARD, HUGO) and provides a close result for (WOW) algorithm when using the database with resized images. On the other hand, the proposed model outperforms many of the state-of-the-art studies in the three algorithms when using the database with the original size. Moreover, the experiments illustrated that training the model with algorithm mismatched dataset can improve the detection accuracy significantly in many cases.


image steganalysis, image steganography, convolution neural network, deep learning, spatial domain, Gabor filter

Subject Categories

Artificial Intelligence and Robotics


© Alaaldin Dwaik

Available for download on Thursday, May 01, 2025

Open Access