Lip-movement based speaker recognition using fusion of canonical angles

A method is presented for authenticating people on the basis of lip movement. It uses the kernel mutual subspace (KMS) method using fusion of canonical angles by kernel Fisher discriminant analysis. Its authentication accuracy is better than that of previously proposed lip-movement authentication methods when the distribution of lip images has a nonlinear structure. The similarity of KMS is canonical angle and some canonical angles are calculated by some principal components. However the method of fusion of canonical angles is not discussed well. So I focused on the method of fusion of canonical angles and proposed the fusion of canonical angles by kernel fisher discriminant analysis. Experimental evaluation of the proposed method demonstrated that it has higher accuracy than other methods based on lip movement.