A frequency domain face recognition technique based on correlation plane features as input to a regression neural network

AbstractIn the area of Face Recognition (FR), the frequency domain approach is a widely used technique. The value of peak-to-sidelobe ratio (PSR) of the correlation peak at the output plane is often used as a measure of recognition. However, the use of such a hard threshold can make a FR system erroneous for recognition and classification. As an alternative to the conventional hard thresholding, three different feature vectors of the correlation plane, e.g. the peak intensity, total energy and PSR, are calculated. These three feature vectors are used to train a Generalized Regression Neural Network (GRNN). The improved False Acceptance Rate (FAR) and consistent Recognition Rate (RR) is reported when tested on the Yale face database