Steganalysis Based on Differential Statistics

• Published in: Cryptology and Network Security pp. 224 - 240
• Main Authors: Liu, Zugen, Ping, Lingdi, Chen, Jian, Wang, Jimin, Pan, Xuezeng
• Format: Book Chapter Conference Proceeding
• Language: English
• Published: Berlin, Heidelberg Springer Berlin Heidelberg 2006; Springer
• Series: Lecture Notes in Computer Science
• Subjects: Applied sciences; Artificial intelligence; co-occurrence matrix; COM; Computer science; control theory; systems; Data processing. List processing. Character string processing; differentiation; Exact sciences and technology; HCF; Memory and file management (including protection and security); Memory organisation. Data processing; Pattern recognition. Digital image processing. Computational geometry; Software; Block code; Spatial data structures; Correlation; Histogram; Statistical analysis; Blindly; Probabilistic approach; Image processing; Image databank; Stéganography; Image compression; Distributed system; Gray scale; Center of mass; Vector support machine; Cryptanalysis; Cryptography; Discrete cosine transforms; Cooccurrence analysis; Computer security; Pattern extraction; Spatial database
• ISBN: 9783540494621; 3540494626
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/11935070_16

Differential statistics were proposed in this paper to disclose the existence of hidden data in grayscale raw images. Meanwhile, differential statistics were utilized to improve the algorithm introduced by Fridrich to attack steganographic schemes in grayscale JPEG images. In raw images, to describe the correlation between data and their spatial positions, co-occurrence matrix based on intensities of adjacent pixels was adopted and the use of co-occurrence matrix was extended to high-order differentiations. The COMs (center of mass) of HCFs (histogram character function) were calculated from these statistics to form a 30-dimensional feature vector for steganalysis. For JPEG files, differential statistics were collected from boundaries of DCT blocks in their decompressed images. The COM of HCF was computed for each of these differential statistics and statistics from DCT domain so that a 28-dimensional feature vector can be extracted from a JPEG image. Two blindly steganalytic algorithms were constructed based on Support Vector Machine and the two kinds of feature vectors respectively. The presented methods demonstrate higher detecting rates with lower false positives than known schemes.