基于递进多目标蛙跳优化的LSB±K隐写算法

A secure LSB plus-minus K steganography based on escalating multi-objective SFLA

设计一种采用非支配排序、拥挤机制和种群重构等策略的递进多目标混合蛙跳算法,可有效保证Pareto前沿的多样性和均匀性.将该多目标优化算法用于设计LSB±K隐写算法.在优化过程中对图像分块,对应所有图像子块组成的矩阵构成优化算法的可行解,矩阵元素代表对应图像块像素的嵌入位数.以载体图像与载密图像差分图的直方图特征函数质心差和隐写容量为两个优化目标,对不同图像块的嵌入位数进行优化.实验结果表明,采用递进多目标混合蛙跳优化的LSB±K隐写算法,与相近抗分析性能下的LSBM隐写及单目标优化LSBM隐写相比,嵌入容量提高了30%;与相同容量的LSB±2隐写及单目标优化LSB±2隐写相比,抗分析能力更强.

An escalating multi-objective shuffled frog leaping algorithm （EMO-SFLA） was proposed. In the EMO-SFLA strategies such as non-dominated sorting, crowding distance and population reconstruction were adopted to obtain a diversity and uniformity of the Pareto frontier. Then, an improved least-significant bit plus-minus K steganography （ denoted as SFLA-LSBK） was put forward based on EMO-SFLA. In SFLA-LSBK, a matrix denoting all the image blocks was defined as a feasible solution, and the value was shown of each element of the matrix, denoting the embedding bits of the corresponding image block. The difference of the histogram characteristic function center of mass （HCF-COM） of the cover difference images and its stego version, together with the embedding capacity, were employed as the two objects for optimizing the embedded bits of each image block. Experimental results show that, under similar security levels, SFLA-LSBK has a 30% increase in embedding capacity over LSB Matching steganography, and the LSB Matching steganography was improved with single objective optimization. The proposed method also demonstrates better performance in resisting steganalysis than LSB ± 2 steganography, and the LSB ± 2 steganography is improved with single objective optimization under the same embedding capacity.