Robust Image Identification without Visible Information for JPEG Images

Abstract

Kenta IIDA† , Student Member and Hitoshi KIYA †a) , Fellow SUMMARY A robust identification scheme for JPEG images is proposed in this paper. The aim is to robustly identify JPEG images that are generated from the same original image, under various compression conditions such as differences in compression ratios and initial quantization matrices. The proposed scheme does not provide any false negative matches in principle. In addition, secure features, which do not have any visual information, are used to achieve not only a robust identification scheme but also secure one. Conventional schemes can not avoid providing false negative matches under some compression conditions, and are required to manage a secret key for secure identification. The proposed scheme is applicable to the uploading process of images on social networks like Twitter for image retrieval and forensics. A number of experiments are carried out to demonstrate that the effectiveness of the proposed method. The proposed method outperforms conventional ones in terms of query performances, while keeping a reasonable security level. key words: image identification, JPEG, social networks IntroductionThe growth of popularity of social networks(SNs) like twitter and Facebook have opened new perspectives in many research fields, including the emerging area of Multimedia Forensics. The huge amount of images have been uploaded on social networks, and generally stored in a compressed form as JPEG images, after being re-compressed with different compression parameters from those of uploaded im-This paper proposes a new robust identification scheme for JPEG images that are generated from the same original image, under various compression parameters, aiming at producing evidence on the integrity of images e.g. tamper detection. The proposed scheme allows not only to identify JPEG images under the use of different initial quantization matrices and quality factors but also to provide no visible information.So far, several identification schemes and image hash functions have been developed for compressed images [6]- [20]. They can be broadly classified into two types according to a difference in extracted features: compression method-depended type, to which the proposed scheme corresponds, and compression method-independent type. not only are robust against a difference in compressed ratios but also do not produce false negative matches in various compression ratios, due to the use of the positive and negative signs of DCT coefficients. However, they need to be combined with a security technique with a secret key such as the fuzzy commitment scheme [14]- [16], to securely protect the features which have some visible information. In addition, the security technique is not useful for the robust identification when various initial quantization matrices and quality factors are used.Due to such situations, this paper proposes a robust scheme for identifying images compressed under the various compression parameters. The strategy to robustly identify JPEG images is to ...

Robust Image Identification with DC Coefficients for Double-Compressed JPEG Images

“…Note that Recall = 100[%] means that there were no false negative matches, and Precision = 100[%] means that there were no false positive matches. It is confirmed that proposed scheme achieved Recall = 100% for all databases, although the conventional compression-dependent schemes [7], [9] did not. The schemes [7], [9] guarantee Recall = 100% for singlecompressed images in principle, however, it does not consider the errors caused by double-compression.…”

“…, the client/user judges that O ′ i and Q ′′ are generated from different original images and the process for image As shown above, by using the feature extracted from DC coefficients, the identification is carried out in the proposed scheme, although the conventional schemes use all DCT coefficients [9]- [11] or the signs of DCT coefficients [7], [8]. The following are the reasons why this feature is used in this paper.…”

“…The proposed scheme was compared with two compression-method-dependent schemes (zero value positions-based scheme [9] and DCT signs-based one [7]) and three image hashing-based schemes (low-rank and sparse decomposition-based scheme [13], quaternion-based one [14] and iterative quantization (ITQ)-based one [15]), where ITQ-based hash values were generated from 512 dimensional GIST feature vectors and each hash value was represented by 512 bits. In the schemes [13]- [15], the hamming distances between the hash value of a query image and those of all images in each database were calculated, and then images that had the smallest distance were chosen as the images generated from the same original image as the query, after decompressing all images.…”

“…On the other hand, due to the use of robust features against JPEG errors, compression-method-dependent schemes [7]- [11] have the stronger robustness than the first type ones. The schemes [7], [8] use positive and negative signs of discrete cosine transform (DCT) coefficients, and the scheme [9]- [11] focuses on the positions in which DCT coefficients have zero values. They guarantee that there are no false negative matches for the identification between single-compressed images.…”

Robust Image Identification for Double-Compressed and Resized JPEG Images

“…On the other hand, due to the use of robust features against JPEG errors, compression-method-dependent schemes [7]- [13] have the stronger robustness than the first type ones. The schemes [7]- [10] use positive and negative signs of discrete cosine transform (DCT) coefficients, and the schemes [11]- [13] focus on the positions in which DCT coefficients have zero values. However, the identification for images having different sizes can not be performed.…”