A subspace-based multi-view face clustering and recognition approach
M. Alarmel Mangai, N. Ammasai Gounden · 2011
In this paper a clustering algorithm has been presented for data sets having faces with large variations in pose. Disjoint clusters are created from low-dimensional subspaces of the data set. Partitioning is carried out in the form of a tree-like structure. The subspace-based linear recognition algorithm, Subclass Linear Discriminant Analysis (SLDA) has been employed for recognizing the faces. The training set for recognition purpose is formed using the group of clusters obtained. The quality of clusters generated by the proposed grouping scheme is compared with the ones generated from K-means clustering algorithm. Experimental results on recognition show that the proposed grouping scheme yields quality clusters compared to K-means.