保局投影在图像隐密检测中的应用

为解决图像隐密检测中图像特征维数过高导致的"维数灾难"问题,在保持图像特征内在低维结构的基础上降低特征向量的维数,方便构造更有效的分类器,提出了一种基于保局投影(locality preserving projections,LPP)降维的图像隐密检测算法,对待测图像进行小波变换形成图像特征后,利用LPP算法实现对图像高维特征的降维,得到图像特征集的低维流形.使用支持向量机(SVM)算法将降维后的特征映射到分类特征空间,实现对正常图像和隐密图像分类.实验结果表明,与不使用降维算法的检测方案相比,基于LPP降维的检测算法能够显著地提高检测的准确率.

Application of Locality Preserving Projections in Image Steganography Detection

Aiming at the "dimensionality curse" problem of image features in image steganography detection,a new detection algorithm based on locality preserving projections is introduced,which is designed to reduce the dimensionalities of the image feature vector while keeping its low dimensional structure at the same time so as to make the classifier more effective .After forming the image feature vector by wavelet transform,the algorithm projects the vector to a low dimensional one and gets the inner manifold by using the LPP algorithm.Support vector machines classifier is used to classify the image set.It is testified that in contrast with the algorithm without dimensionality reduction,this algorithm can improve the performance signature signally.