摘要
为克服主成分分析(PCA)、线性判别分析(LDA)等经典人脸识别方法在处理人脸数据时采用的向量化破坏数据间的内在结构这一不足,将张量思想运用到稀疏表示理论中,提出一种基于加权核Gabor特征的张量稀疏人脸识别算法。算法选取Gabor特征作为人脸识别的研究数据,同时用高斯核方法将数据映射到高维空间,这一应用保持了图像信号在局部时间和局部频带上的频谱信息,减小了破坏数据结构所造成的误差;而张量散度思想的运用保持了数据间的区域几何结构。在ORL、YALE和AR人脸数据库上的仿真实验表明了该方法的识别效果,且在光线变化、姿态变化以及训练样本数不足情况下具有较好的鲁棒性。
To avoid the shortage that the intrinsic structure of face data is destroyed when vectorization is used in PCA (Principal Component Analysis), LDA (Linear Disereminant Analysis) and other classic face recognition approaehs, the tensor idea was applied to the sparse representation theory, and a tensor sparse face recognition algorithm based on weighted kernel Gabor features was proposed. The Gabor feature was selected as research data and Gauss kernel was used to map data to high-dimensional space, keeping the spectral information of image signals on the local time and local bands, reducing the error caused by the destruction of data structure; the tensor divergence was adopted to keep the regional geometrical structure between data. The simulation experiments show that the proposed method is effective to the face recognition on ORL, YALE and AR datasets, and is robust to lighting, pose change and insufficient samples.
出处
《计算机应用》
CSCD
北大核心
2016年第A02期185-188,536,共4页
journal of Computer Applications
基金
国家自然科学基金资助项目(61572087)