Statistical Assessment of PCA/SVD and FFT-PCA/SVD on Variable Facial Expressions
Louis Asiedu *
Department of Statistics, University of Ghana, Legon-Accra, Ghana.
Atinuke O. Adebanji
Department of Mathematics, Kwame Nkrumah University of Science and Technology, Kumasi, Ghana.
Francis T. Oduro
Department of Mathematics, Kwame Nkrumah University of Science and Technology, Kumasi, Ghana.
Felix O. Mettle
Department of Statistics, University of Ghana, Legon-Accra, Ghana.
*Author to whom correspondence should be addressed.
Abstract
Face recognition is a dedicated process in the human brain. Automatic face recognition is rewarding since an efficient and resilient recognition system is useful in many application areas. Recent face recognition algorithms are still faced with the challenge of recognizing face image under variable environmental constraints. This paper presents a statistical evaluation of the performance of two face recognition algorithms namely, Principal Component Analysis with Singular Value Decomposition (PCA/SVD) and Principal Component Analysis with Singular Value Decomposition using Fast Fourier Transform for preprocessing (FFT-PCA/SVD) on variable facial expressions (Angry, Disgust, Fear, Happy, Sad and Surprise) along with their neutral expressions. We considered 42 individuals from Cohn Kanade Facial Expressions database, Japanese Female Facial Expressions (JAFFE) and a created Ghanaian Face database for recognition runs. Multivariate statistical methods were used in the assessment of the face recognition algorithms. GNU Octave was used to perform all numerical runs and statistical evaluation of the recognition algorithms. The results of the statistical evaluation show that, FFT-PCA/SVD is comparatively consistent (Low variation) and efficient (Higher recognition rate) than PCA/SVD algorithm in the recognition of variable facial expressions. The paper also proposes Fast Fourier Transform as a viable noise removal mechanism that should be adopted during image preprocessing.
Keywords: Fast fourier transform, multivariate, principal component analysis, singular value decomposition.