混合蛙跳优化决策面的LSB±k隐写算法

Improved LSB±k Steganography Based on Decision Surface Optimized by SFLA

针对现有智能优化改进隐写不能对高维特征同时进行优化的问题,提出了一种混合蛙跳优化决策面的改进LSB±k隐写算法(记为SFLA-LSB±k).不同于其他优化改进隐写中尽可能减少图像载密前后某种特征变化的策略,在SFLA-LSB±k中,通过优化载密图像的特征变化,使载密图像特征变化方向随机化,导致分类器无法训练出一个能对载体与载密图像进行分类的决策面,从而达到抵抗分析的目的.实验结果表明,与标准的LSB±k隐写和相关PSO优化改进LSB±k隐写相比,SFLA-LSB±k有效提高了LSB±k的安全性,特别是当k取1时,该算法针对78维特征隐写分析的AUC值可下降到0.563 7.

Most improved steganographies based on intelligent optimization cannot realize high-dimensional features optimization simultaneously. To solve the problem, this paper proposes an improved LSB±k steganography （denoted SFLA-LSB＋k） based on SFLA to achieve an optimal decision surface. Different from the other improved steganographies that attempt to keep image features unchanged after data embedding as much as possible, the proposed method tries to randomly change feature directions of stego-image in the embedding process optimized by SFLA. Thus, it is difficult to find a decision surface to distinguish cover images from stego-images. Simulation indicates that, with the same em- bedding capacity, SFLA-LSB=hk demonstrates better performance in resisting steganMysis than the traditional LSB＋k and the improved LSB＋k optimized by PSO. Especially, theAUC value is reduced to 0.563 7 when k=l against steganalysis with 78-dimension features.