Steganography detection feature selection method based on MMD residual error

Abstract

The invention provides a steganography detection feature selection method based on MMD residual errors. The method comprises the following steps of: 1, disassembling a high-dimensional Rich Model steganography detection feature into a plurality of Rich Model sub-model feature vectors; 2, for each Rich Model sub-model feature vector, measuring an attribute importance value of each feature component based on an MMD residual attribute importance measurement formula, and carrying out descending sorting on each feature component according to the attribute importance value; 3, for each Rich Model sub-model feature vector, setting a window size Spro, and selecting a feature component of a front Spro dimension in the vector after descending sorting as a steganography detection feature of the current Rich Model sub-model feature vector after dimension reduction; 4, combining the feature vectors of the Rich Model sub-models after dimension reduction to serve as final steganography detection features. According to the method, the feature dimension of Rich Model steganography detection can be efficiently reduced, and the original detection performance is maintained.

Description

technical field [0001] The invention relates to the technical field of steganography detection, in particular to a feature selection method for steganography detection based on MMD residuals. Background technique [0002] Since Fridrich et al. proposed the HUGO steganography algorithm based on the "distortion function design + STC coding" framework in 2010, the adaptive steganography algorithm has gradually become the mainstream image steganography algorithm. These algorithms overshadow traditional steganographic detection features. In the process of fighting against adaptive steganography, extracting high-dimensional steganographic detection features has become the main means of detecting adaptive steganography. Among the high-dimensional steganographic detection features, the Rich Model structure is the most typical. Based on this structure, researchers have successively proposed steganographic detection features such as SRM, CCJRM, PHARM, PSRM, DCTR, and GFR. The introd...

AI Technical Summary

Problems solved by technology

In paper 1 (“G.R.Xuan, X.M.Zhu, P.Q.Chai, Z.P.Zhang, Y.Q.Shi, D.D.Fu. Feature selection based on the Bhattacharyya distance. In Proc. the 18th International Conference on Pattern Recognition, Hong Kong, China, Aug. 20-24 , 2006, pp.1232-1235") and text 2 ("L.D.Jennifer, J.Jaikishan. Feature selection for steganalysis using the Mahalanobis distance. In Proc. the IS&T-SPIE Electronic Imaging Symposium on MediaForensics and Security II, San Jose, CA, USA, January 18-20, 2010, SPIE 7541, pp.754104-1-754104-12"), researchers proposed a method to measure the importance of single-dimensional feature component attributes based on Bhattacharyya distance and Mahalanobis distance, and selected the carrier The feature vector with a large distance between the image feature and the secret image feature is used for steganography detection, but the effect of dimensionality reduction is not outstanding

Images