Expression-Invariant 3D Face Recognition Using K-SVD Method

This paper proposes a method to perform expression invariant face recognition using dictionary learning approach. The proposed method performs the operation in the following stages: the T-region extraction from the face to get the facial region having minimum variation with expression, determination of the wavelet coefficients of the extracted region, dictionary learning using K-SVD and matching. The experiment has been performed on a database that contains 40 persons with 9 expressions each under different illumination conditions. The recognition performed has shown a good accuracy rate as compared to the mostly used PCA-SVM approach. Our system uses label-consistent K-SVD algorithm for dictionary learning to learn a set of dictionaries that represents 3D information of the face. This method fulfills the purpose of sparse coding and classification.