Face recognition using the wavelet approximation coefficients and fisher's linear discriminant
Lin Cao, Dengyi Chen, Jing Fan · 2012
This paper introduces a face recognition method using Fisher's linear discriminant in the Wavelet domain composed of the Wavelet approximation coefficients (WAFLD). As opposed to other approaches for face recognition, the proposed method makes use of the approximation coefficients matrices obtained by three-level Wavelet decomposition of the input image, and the new image matrix is reshaped by combination of the approximation coefficients. Subsequently, the Fisher's linear discriminant is applied to the new image matrix for face recognition. The feasibility of the new WAFLD method has been successfully tested on face recognition using ORL and 1200 CAS-PEAL-R1 frontal face images corresponding to 200 subjects, which were acquired under variable illumination and facial expressions. The novel WAFLD method achieves 99% accuracy on face recognition using only 20 features.