Steganalysis of LSB Matching Exploiting High-Dimensional Correlations between Pixel Differences

In this paper, a new steganalytic method exploiting high-dimensional correlations between pixel differences is proposed, which is designed to detect the presence of spatial LSB matching steganography in grayscale images. The proposed steganalytic method is based on a Fisher LDA(Linear discrimination analysis) classifier trained on feature vectors corresponding to cover-images and stego-images,and the distinguishing features are built from marginal and joint statistics of pixel differences. Experimental results show that the proposed method exhibits excellent performances for the detection of LSB matching steganography in grayscale images. Moreover, it has a low computational complexity and fast computational speed.The proposed steganalytic method can also be applied to the detection of other spatial-domain steganography methods,and the idea can be used for reference while designing steganalytic methods of JPEG steganography.