Skip to Main Content
Face Recognition is the process of identification of a person by his facial image. As applied to face recognition, this paper proposes a method, comprising of Laplacian of Gaussian (LoG) filter for intricate facial detail enhancement, Singular Value Decomposition (SVD) for holistic feature extraction and Feed forward Neural Network (FFNN) for classification. Applications of LoG filter highlights, otherwise hidden details such as wrinkles, moles, etc. The principal components from SVD form a basis for the original dataset. The original dataset is then projected onto the reduced subspace, the result of which is weight vectors that are fed as input to the FFNN for the training task. The FFNN uses gradient descent batch mode, back propogation algorithm with adaptive supervised learning. The method is christened L-SVD-NN and is tested on the Yale face dataset achieving an accuracy of 84.85%. This method achieves an added advantage of reduction in dimensionality.