Steganalysis of JPEG Images based on 2D-Gabor Filters and Feature Dimensionality Reduction

when steganography changes are limited to regions of complex texture, the JPEG stenographic methods struggle to preserve image texture features in all orientations and scales. As a result, a 2D-Gabor filter and PCA-based steganalysis feature extraction method is proposed. The 2D-Gabor filters are sensitive to the number of ideal joint localization characteristics in both the spatial and frequency domains. Specifically, their ability to analyze the texture features of an image at varying scales and orientations makes them more effective at detecting the artifacts brought on by steganography embedding. For the suggested technique, various 2D-Gabor filters are created, which benefit the scale and orientation sensitivity analysis and are used to filter the JPEG image. Then, extract the features from the filtered image and reduce its dimensionality by averaging according to the symmetries of Gabor filter orientations. After that, the final feature vector is constructed by combining all the features. Furthermore, the PCA method reduces the feature dimension and improves the classification performance. Finally, the SVM classifier receives the resultant feature vector to decide between the “normal” and “abnormal” image. Experiments demonstrate that the proposed approach for image steganalysis can achieve competitive accuracy compared with the DCTR and PHARM steganalysis features and with the extracted feature without applying the Gabor filters.