To solve the problems of that principal component analysis (PCA) is sensitive to light and affected its recognition rate by external interference factors, this paper puts forward an improved algorithm which based on the fusion of PCA and LDA. Firstly, the input face image is preprocessed with eliminating noise, normalization and gray level distribution equalization. Then using PCA algorithm to project the face training image to obtain low-rank feature subspace, and then the subspace face feature is derived from LDA, thus the fusion feature space is acquired. Finally the training and test samples are projected to the fusion feature space, and identify test samples based on the nearest neighbor rule. The experiment shows that this algorithm can fuse the advantage of PCA and LDA effectively and improve the robustness and efficiency of the system.
Paper
Full text
Research on Improved PCA Face Recognition Algorithm
Semantic Scholar · Computer Science · 2019
Abstract
To solve the problems of that principal component analysis (PCA) is sensitive to light and affected its recognition rate by external interference factors, this paper puts forward an improved algorithm which based on the fusion of PCA and LDA. Firstly, the input face image is preprocessed with eliminating noise, normalization and gray level distribution equalization. Then using PCA algorithm to project the face training image to obtain low-rank feature subspace, and then the subspace face feature is derived from LDA, thus the fusion feature space is acquired. Finally the training and test samples are projected to the fusion feature space, and identify test samples based on the nearest neighbor rule. The experiment shows that this algorithm can fuse the advantage of PCA and LDA effectively and improve the robustness and efficiency of the system.