基于MB_LBP与改进Fast PCA算法的人脸特征提取

Facial feature extraction based on MB_LBP and improved Fast PCA algorithm

针对MB_LBP算法对人脸特征提取维数较高,使用PCA方法会造成图像原始空间结构破坏和维数变得过大等问题,提出一种基于多块LBP(Multi-scale Block Local Binary Patterns,MB_LBP),结合改进的Fast PCA算法进行人脸特征提取的方案。首先用MB_LBP算法提取人脸图像的特征,接着用本文所改进Fast PCA方法加速计算矩阵S非零本征值所对应的本征向量,对人脸特征进行降维,最后在ORL人脸库进行验证。实验表明,该方法对后期人脸特征提取效果优于改进前的效果,很大程度上降低了提取时间,效果明显。

MB_LBP algorithm for facial feature extraction will cause higher dimensions and using the PCA method can cause structural damage to the image of the original space and larger dimension. So a multi-block LBP（Multi-scale Block Local Binary Patterns,MB_LBP）, combined with the improved Fast PCA algorithm for face feature extraction is proposed. First using algorithm MB_LBP to exlract face image feature, then with the improved Fast PCA algorithm to accelerate compute eigenvalues nonzero matrix S corresponding eigenvectors for facial feature dimensionality reduction. And finally it is verified in the ORL face database. Experiments show that the method has bettter effects for later facial feature extraction than before. It largely reduces the extraction time. The effects are obvious.