|
Abstract:
|
A new approach to develop facial recognition using principal component analysis of interval-valued data is proposed. We exploit patterned covariance structures in doing so and we accomplish this in two stages: first, we get eigenblocks and eigenmatrices of the variance-covariance matrix, and then we analyze these eigenblocks and the corresponding principal vectors together in some seemly sense to get the principal components of the interval-valued data. We apply our method to the face recognition data in Douzal-Chouakria (in Statistical Analysis and Data Mining 4(2): 229-246, 2011, Table 1). We study this dataset as 9 independent faces with 3 repeated interval-valued faces for each face in contrast to 27 independent faces considered by the previous authors. We take care of the 3 repeatedness of each face by using structured covariance matrices and answer the question whether three sequences (3 repeatedness) belong to the same face or not. Face sequence recognition or classification is an important problem as face might slightly change due to several reasons. Results illustrating the accuracy and appropriateness of the new method over the existing methods are presented.
|
