A feature based technique for face recognition using Mexican hat wavelets

A new features based algorithm for face recognition is proposed. The algorithm is based on locating features on faces and finding the correspondence of these features between faces using an algorithm which is adopted from the works of Yasein and Agathoklis. The algorithm is using feature extraction through filtering with Mexican-hat wavelets and detection of correspondence of the feature points through comparison of Zernike moments. The transformation of the feature points between images is obtained using an iterative weighted least-squares algorithm which allows to eliminate outliers. The face recognition is based on the residual error between the feature points after transformation. The performance of the proposed method is evaluated using experiments with a face database and compared against existing techniques for face recognition. The results of experiments with varying pose and illumination conditions show that the proposed method performs well compared to existing feature based face recognition techniques.