Face recognition by fiducial point analysis

In this paper we present a hybrid method for face recognition task, which covers both the important information in the high order relationships among the image pixels and low dimensional representation of the image. This system is feature based, and analyzes local facial features, which are located by a meta-version of the sparsification algorithm in the context of LFA (local feature analysis) technique. The LFA sparsification strategy reduces the dimensionality of the representation by choosing a subset of output points that are as decorrelated as possible. Since choosing different set of points for each image, will make a problem in recognition, we determined a single set of points for all images which called fiducial points, and propose a method based on genetic algorithm to find them in any face image. After extracting fiducial points, we describe each of them with a set of Gabor coefficients to make a unique ID for each new face.