Benchmarking for steganography by kernel fisher discriminant criterion

Abstract

In recent years, there have been many steganographic schemes designed by different technologies to enhance their security. And a benchmarking scheme is needed to measure which one is more detectable. In this paper, we propose a novel approach of benchmarking for steganography via Kernel Fisher Discriminant Criterion (KFDC), independent of the techniques in steganalysis. In KFDC, besides between-class variance resembles what Maximum Mean Discrepancy (MMD)merely concentrated on, within-class variance plays another important role. Experiments show that KFDC is qualified for the indication of the detectability of steganographic algorithms. Then, we use KFDC to illustrate detailed analysis on the security of JPEG and spatial steganographic algorithms. © 2012 Springer-Verlag Berlin Heidelberg.