Therefore, if one wants to extract the IDSF of F5-like steganography distinguished from others, it is important to construct feature extraction sources according to the coefficients under case 2 and case 3 in Eq.
17) and (18), the difference histogram will be calculated in this paper and taken as IDSF of F5-like steganography.
At last, the IDSF of F5-like steganography extracted from
In summary, based on the analysis of modifications and influences caused by F5-like steganography to the data in DCT domain and spatial domain, the IDSF of F5-like steganography distinguished from others can be obtained as follows:
According to the IDSF extraction methods proposed in Section 3 for F5-like steganography, extract features of the experimental stego images set.
In addition, for the reason that the detectors may not know clearly the steganographic algorithms possibly used in MultiClsStegImgs in practical steganalysis, therefore, this paper will analyze and test the performance of the proposed IDSF based recognition algorithm in the condition that only one type of steganography is known to the detector, which is used to recognize F5-like stego images.
When some of the steganographic algorithms possibly used in the multi-class stego images set are unknown, based on the proposed IDSF of F5-like steganography and the feature in , the classification and recognition procedure presented in Subsection 3.
2 for recognition of F5-like stego images, the recognition results based on the proposed IDSF are superior to that based on the CCPEV feature in .
The results in Table 4 and Table 5 indicate further that, when only the steganographic algorithm JSteg is known, that is to say, the Steghide and Outguess stego images are not included in training classifier and are unknown to the classifier, the average detection accuracy and time consuming based on the proposed IDSF F5likeIdf for F5-like steganography are obviously superior to that based on the existing feature CCPEV.