中文摘要：

如何降低高维数据的维数而不损失原有数据的内在信息是机器学习和数据挖掘领域中的热点问题．本文在图嵌入框架的基础上提出一种新的降维分析算法IKLDA（improved kernel Linear discriminant analysis），不仅使得隐藏在图像的信息能被区分出来，而且大大降低了数据的维数，理论分析及实验结果表明IKLDA的降维隐写分析是有效的，比其它传统降维方法效果要好，并且进一步推进了数据挖掘可视化方法在隐写分析的应用．

英文摘要：

Reducing the dimensionality of data without losing intrinsic information is a hotspot in machine learning and data mining. In this paper we propose a new dimensionality reduction algorithms call IKLDA（improved kernel Linear discriminant analysis） on the ground of graph embedding framework. Our method not only can detect the information hidden in digital images but also reduce the dimensionality. Theoretical analysis and experiments show that our new IKLDA algorithm is effective in steganalysis and is more precise than the other traditional dimensional reduction methods. Furthermore, our method promotes development of visualization in the application in steganalysis.