Title :
Locality Preserving Projections Based on L1 Graph
Author :
Liu, Zhonghua ; Yin, Jun ; Jin, Zhong
Author_Institution :
Sch. of Comput. Sci. & Technol., Nanjing Univ. of Sci. & Technol., Nanjing, China
Abstract :
Traditional graph construction methods usually decompose the graph construction process into two steps and have certain parameters which required manual setting. In addition, since changes in pose, illumination and expression would cause large changes in the appearance, this often leads to the fact that neighboring samples don´t belong to the same class. However, these faults are overcome by the LI graph construction method which is different from conventional graph construction method. In LI graph method, the graph adjacency structure and the graph weights are derived simultaneously in a parameter-free manner. Combined with LI graph method and locality preserving projection method, the novel method called locality preserving projection based on LI graph is proposed. Experimental results on face databases demonstrate the effectiveness of our method.
Keywords :
graph theory; lighting; pose estimation; visual databases; L1 graph construction; face databases; illumination; locality preserving projections; parameter-free manner; sparse representation; Databases; Face; Face recognition; Lighting; Nearest neighbor searches; Sparse matrices; Training;
Conference_Titel :
Pattern Recognition (CCPR), 2010 Chinese Conference on
Conference_Location :
Chongqing
Print_ISBN :
978-1-4244-7209-3
Electronic_ISBN :
978-1-4244-7210-9
DOI :
10.1109/CCPR.2010.5659156