利用LLE重建的人脸图像多角度转化算法  

Multiview Face Image Transformation Algorithm Using LLE

作　　者：马祥[1] 宋焕生[1] 

机构地区：[1]长安大学信息工程学院,陕西西安710064

出　　处：《电视技术》2013年第19期50-52,共3页Video Engineering

基　　金：国家自然科学基金项目(61101215)

摘　　要：针对现有多角度人脸图像相互转化算法存在复杂度高、运算量大、重建结果中头发、脸部轮廓部位比较模糊的问题,提出一种利用局部性嵌入法(LLE)进行重建的多角度人脸图像相互转化算法。将人脸特定角度空间的重建系数运用到目标角度空间,在权值求取时运用局部线性嵌入非线性降维算法,并将算法中的局部协方差矩阵进行大常数对角加载。对比实验结果表明,该算法简单,计算速度快,转化后图像质量高,并且在头发和人脸边缘部分合成效果明显优于现有算法。Most existing multi-view face image transformation methods are of high complexity,large amount of computation,and the parts of hair and facial contour are rather blur in reconstruction results.This paper proposes a multi-view face image transformation algorithm based on locally linear embedding（LLE）.It applies the space cc efficients of certain face view to the target view,and the weights are obtained using LLE,which is a nonlinear dimensionality reduction algorithm.Furthermore the local covariance matrix in the algorithm needs large constant diagonal loading.The Compared experimental results show that the algorithm is simple,fast in calculation,the transformed image is of high quality,and synthetic effect of the edge part of the hair and the face are more superior than some existing algorithms.

关 键 词：人脸图像 局部线性嵌入 非线性降维 多角度转化 

分 类 号：TN911.73[电子电信—通信与信息系统]