Skip to Main Content
This study presents a novel combination of Radon transform and linear and kernel PCA methods for high performance face recognition. Radon transform is well known in image processing due to its simplicity and invariance to rotation. It's discrete version is used to extract a number of characteristic features from 2-D facial images through taking discrete Radon transform over a set of angular directions. The resulting Radon transform features are projected into a lower dimensional space using principal component analysis through which principal components of the extracted features are determined. Finally, these principal components and a simple Euclidean distance measure are used for face recognition. Experimental evaluations over the well-known FERET database demonstrated that quite significant improvements are achieved from the hybridized Radon transformation and PCA approaches.