Abstract:
Face Recognition is the most common and popular technique in the current authentication bio metric systems. It is the most acceptable because it can be used remotely without the collaboration with object. However, the method of face recognition is sensitive to the variation of the lighting and the change of the position of the face during the acquisition of the image. Face authentication systems typically use the grayscale representation of the face image as an input feature of these systems. But we propose a new method based on the one-dimensional statistics of the face image and the use of the color representation that improves the performance of these systems. We tested several color spaces for the transformation of the RGB colorimetric components of the original images. The results obtained in the different spaces, colorimetric components are combined by the use of a nonlinear fusion for classification with a single neuron network of RBF type. A new technique for face authentication called MS has been proposed which gives a 94.75% success rate with the nonlinear fusion of the colorimetric components of the YCrCb color space and using grayscale 90.53% success rate is achieved. This is an improvement of 04% in the rate of success compared to the use of grey scale images. The results presented show the interest of the development of the new approach which makes it possible to reduce the computing time thanks to its simplicity and the robustness when working with a large database and that the color information increases the performance of this face authentication system. To validate this work, we tested these approaches on frontal images of the XM2VTS database according to its associated Lausanne protocol.
