An efficient wavelet/neural networks-based face detection algorithm

In this paper, we proposed an efficient method to address the problem of face detection that is based on neural networks (NNs) and wavelet representation. We utilized a multi-layer perceptron (MLP) so as to classify skin and non-skin pixels in the YCrCb color space. In this work, skin samples in images with varying lighting conditions are used to obtain a wide skin color distribution. The training data is generated from positive and negative training patterns in the Cb-Cr planes. Subsequently, training set is fed to an MLP, trained using the Levenberg-Marquardt algorithm using these skin samples. We apply the above mentioned NN-based skin classifier to the chrominance values corresponding to the coarsest level of the chrominance approximation subimages obtained from wavelet transform to classify the candidate face pixels. Furthermore, we have proposed a subspace approach in the space-frequency domain for the fast detection of face utilizing wavelet representation.