Anti-forensics of JPEG compression detection schemes using approximation of DCT coefficients

Abstract

Here we propose a unique anti-forensics method using the approximation of DCT coefficients for security analysis of forensics schemes that are dependent on the distortions produced by JPEG compression to detect forgery. Approximation process first builds a model of the AC component of DCT coefficients. Then it uses that model to restore the values of the DCT coefficients. The beauty of the proposed anti-forensics technique is that one can specify any distortion measure and it is capable of minimizing that distortion as long as the specified distortion is produced by JPEG compression or the distortion is measurable in the DCT domain. We specifically mount our anti-forensics technique on three leading JPEG compression detection schemes (Fan and de Queiroz, IEEE Trans Image Process 12(2):230–235, 2003; Lai and Bohme 2011) to highlight their weaknesses. Though these schemes are based on three different distortions metrics, still we could achieve $100\%$ missed detection rate for JPEG images having quality factor more than equal to $60\%$ . Our analysis raises a serious question regarding the robustness and security of JPEG artifact based forgery detection schemes.