Universal Steganalysis for image Based on Genetic Algorithm and Grey-SVC

The isolated samples can produce some effect on distinguishing the best classifying plane, which becomes one of causes of less performance of universal steganalysis that uses Support Vector Machines (SVM) as classifier. This paper proposes a new universal steganalysis algorithm for image based on Genetic Algorithm (GA) and Grey Support Vector Machines (GSVM). The algorithm firstly catches characteristic of noise signal in wavelet domain of image, then utilizes GA search samples which are used to train, and finds the best characteristic of species, finally makes grey relational degree between sample characteristic and the best characteristic of species participate in training of SVM, thus constructs a GSVM to be a classifier of steganalysis. The result testing on the large numbers of images indicates that the proposed universal steganalysis algorithm has less false positive rate and better classifying performance compared to Holotyak’s algorithm which has the same characteristic with above algorithm, which indicates that GSVM can reduce effect of isolated samples.