Moments and Similarity Measure Feature Based Image Steganalysis Technique (MSM)

Steganalysis is art and science of detecting messages hidden using steganography. In this article a novel universal image steganalysis technique is proposed in which various moments like invariant, statistical moments and Zernike and various other similarity measure parameters like MSE, PSNR, RMSE, SSIM, Shannon’s ENTROPY, DC Coefficient, DCT distance, NAE and Average Difference i.e. total 24 features are selected as the features. The proposed steganalysis methods has been designed based on three classifiers i) Back Propagation Neural Network and ii) SVM classifier and iii) K-nearest neighbor classifier. The proposed universal steganalyzer has been tested against few of the various well-known steganographic techniques which operate in both the spatial and transform domains. The experiments are performed using a large data set of JPEG and BMP images obtained from publicly available websites. The image data set is categorized with respect to different features of the image to determine their potential impact on steganalysis performance. Experimental results demonstrate the effectiveness and accuracy of the proposed technique compared to other existing techniques.