Single Image Face Recognition Using Laplacian of Gaussian and Discrete Cosine Transforms

This paper presents a single image face recognition approach called Laplacian of Gaussian (LOG) and Discrete Cosine Transform (DCT). The proposed concept highlights a major concerned area of face recognition i.e., single image per person problem where the availability of images is limited to one at training side. To address the problem, the paper makes use of filtration and transforms property of LOG and DCT to recognize faces. As opposed to conventional methods, the proposed idea works at pre-processing stage by filtering images up to four levels and then using the filtered image as an input to DCT for feature extraction using mid frequency values of image. Then, covariance matrix is computed from mean of DCT and Principal component analysis is performed. Finally, distinct feature vector of each image is computed using top Eigenvectors in conjunction with two LOG and DCT images. The experimental comparison for LOG (DCT) was conducted on different standard data sets like ORL, Yale, PIE and MSRA which shows that the proposed technique provides better recognition accuracy than the previous conventional methods of single image per person i.e., (PC)2A and PCA, 2DPCA, B-2DPCA etc. Hence with over 97% recognition accuracy, the paper contributes a new enriched feature extraction method at pre-processing stage to address the facial system limitations.