复杂模式网络多优先级隐写信息标记仿真研究

Research on Multi-priority Stealth Information Marker Simulation in Complex Pattern Networks

当前多优先级隐写信息标记方法忽视了标注集整体关联性对标记结果的影响,导致标注集整体关联性缺乏,标记结果准确性不高。为了解决上述问题,提出基于标注词精化的复杂模式网络多优先级隐写信息标记方法。针对复杂模式网络噪声干扰,提出标注词精化思想,采用优先级排序特征强化隐写信息重要性权重,完成第一次标注词精化;考虑到各优先级标注集整体关联性对标记结果的影响,计算隐写信息的多优先级相关度评分,进一步优选可以准确刻画隐写信息内容的标题,完成第二次标注词精化。结合多核学习方法优化各优先级标注词特征的相应权重,同时确定分类器的参数,采用训练好的多核支持向量机分类器对不同优先级特征进行分类,标记不同优先级的隐写信息。实验结果表明,所提方法获得的标注词能够较好地描述不同优先级下的隐写信息,标记性能提升明显。

In order to solve the problem about the lack of overall relevance of tagging set and the low accuracy of tagging result,a method to tag the multi-priority steganography information in complex pattern networks based on refinement of annotation word was proposed. At first,the priority ranking feature was used to enhance the weight of importance of steganography information and complete the first refinement of tagging word. Considering the influence of overall relevance of each priority tagging set on the tagging results,this research calculated the score of multi-priority correlation degree of steganography information,so as to further select the title which could accurately describe the content of steganography information. Thus,the second refinement of tagging word was completed. Combined with the multiple kernel learning method,corresponding weight of feature of each priority tagging word was optimized. Meanwhile,the parameters of classifier were determined. Finally,the trained multiple kernel support vector machine classifier was used to classify the different priority features and tag steganography information with a different priority.From simulation results,we can see that the annotation words obtained by the proposed method can describe the steganography information under different priorities well. Meanwhile,the tagging performance is improved obviously.