An Angular Spatial Linear Dimensionality Reduction Face Recognition Algorithm for Real Scenes
Download as PDF
DOI: 10.23977/cii2019.23
Author(s)
Lili Jin, Ying Huang
Corresponding Author
Ying Huang
ABSTRACT
Since the original image sample is small and the feature dimension is high, the effective recognition of the face has always been a difficult point. In order to solve the problem of parameter selection in local projection, and rationally use sample label information to enhance the discriminativeness of sample features after dimensionality reduction, this paper proposes a local Fisher criterion discriminant projection algorithm for angular spatial linear dimension reduction for real scenes. The algorithm adaptively selects the neighbor's neighbor parameter K to make the distribution relationship between samples as true as possible. By constructing the local Fisher criterion to discriminate the projection objective function, the similar samples can be better represented by the same dimension after the dimensionality reduction of the projection, and the samples of the same kind are not significantly different. Finally, the experiment was performed on the Yale face database. The results show that the proposed algorithm can effectively achieve dimensionality reduction and has a higher face recognition rate.
KEYWORDS
Face recognition, Linear dimensionality reduction, Discriminant projection