Robust Image Identification for Double-Compressed JPEG Images (1807.06928v1)

Abstract: It is known that JPEG images uploaded to social networks (SNs) are mostly re-compressed by the social network providers. Because of such a situation, a new image identification scheme for double-compressed JPEG images is proposed in this paper. The aim is to detect single-compressed images that have the same original image as that of a double-compressed one. In the proposed scheme, the signs of only DC coefficients in DCT coefficients and one threshold value are used for the identification. The use of them allows us to robustly avoid errors caused by double-compression, which are not considered in conventional schemes. The proposed scheme has applications not only to find uploaded images corresponding to double-compressed ones, but also to detect some image integrity. The simulation results demonstrate that the proposed one outperforms conventional ones including state-of-art image hashing one in terms of the querying performance.