TY - JOUR TI - The Gaussian Orthogonal Laplacianfaces Modelling in Feature Space for Facial Image Recognition AU - Muntasa, Arif IS - Vol 18, No 2 (2014) PB - Directorate of Research and Community Services, Universitas Indonesia JO - Makara Journal of Technology PY - 2014 SP - 79 EP - 85 UR - http://journal.ui.ac.id/technology/index.php/journal/article/view/2946 AB - Dimensionality reduction based on appearance has been interesting issue on the face image research fields. Eigenface and Fisherface are linear techniques based on full spectral features, for both Eigenface and Fisherface produce global manifold structure. Inability of them in yielding local manifold structure have been solved by Laplacianfaces and further improved by Orthogonal Laplacianfaces, so it can yield orthogonal feature vectors. However, they have also a weakness, when training set samples have non-linear distribution. To overcome this weakness, feature extraction through data mapping from input to feature space using Gaussian kernel function is proposed. To avoid singularity, the Eigenface decomposition is conducted, followed by feature extraction using Orthogonal Laplacianfaces on the feature space, this proposed method is called Kernel Gaussian Orthogonal Laplacianfaces method. Experimental results on the Olivetty Research Laboratory (ORL) and the YALE face image databases show that, the more image feature and training set used, the higher recognition rate achieved. The comparison results show that Kernel Gaussian Orthogonal Laplacianfaces outperformed the other method such as the Eigenface, the Laplacianfaces and the Orthogonal Laplacianfaces.