An efficient approach to rotation invariant face detection using PCA, generalized regression neural network and Mahalanobis distance by reducing search space

This paper introduces a new effective method for human face recognition, which employs Mahalanobis distance, a color segment method, to reduce the search space for a face in an image. The color segment method is used to detect the area where a face may exist and searching for a face by the network is localized in that area. The face, with its orientation, is recognized using principle components analysis (PCA), generalized regression neural networks (GRNN) and Mahalanobis distance. The idea behind PCA is to reduce the dimension of the input vectors and extracting a feature vector. GRNN used as a function approximation network to detect whether the input image contains a face or not and if exists then reports about its orientation. The proposed system shows how GRNN can perform better than back-propagation algorithm and give some solution for better regularization. The proposed classifier system demonstrates advantages in: 1) better capability of approximation to underlying functions, 2) faster learning speed and 3) high robustness to outliers. The proposed technique overcomes one of the main problems for searching method reducing the computing time in two ways; firstly reducing the area to be searched and secondly reducing the scaling factor of the selected area of the image.