A New Robust and Discriminating Method for Face Recognition Based on Correlation Technique and Independent Component Analysis Model

We demonstrate a novel technique for face recognition combined the independent component analysis (ICA) model with the optical correlation technique. Our approach relies on the performances of a strongly discriminating optical correlation method along with the robustness of the ICA model. Simulations were performed to illustrate how this algorithm can identify a face with images from the Pointing Head Pose Image Database (PHPID). While maintaining algorithmic simplicity, this approach based on ICA representation significantly increases the true recognition rate compared to that obtained with an all numerical ICA identity recognition method, that we recently developed, and with another based on optical correlation and a standard composite filter. 2 The field of face recognition has made significant progress in recent years. Computational models of face recognition are important because they can contribute to theoretical insights, e.g. image codes, as well as practical applications, i.e. biometrics, security systems, and human-computer interaction. Many different implementations are being actively explored, including eigenfaces [1], Gabor wavelets [2], and principal component analysis (PCA) [3]. Noteworthy, the independent component analysis (ICA) model has sparked interest in searching for a linear transformation to express a set of random variables as linear combinations of statistically independent source variables [4]. ICA provides a more powerful data representation than PCA as its goal is that of providing an independent rather than uncorrelated image decomposition and representation. Interest in ICA mixture models is motivated in part by their exceptional properties, including high sensitivity to high-order relationships among pixels and enhanced face recognition performance when it is carried out in a compressed and whitened space [5]. Motivated by these developments, we investigate an optical recognition technique using ICA as a preprocessing stage for face recognition. In this Letter, our two primary goals are to validate the principle of the new algorithm proposed to recognize a target face using reference facial images contained in the (supervised) learning base, and to focus on its implementation using a standard correlator which can be optically or numerically implemented. In this paper, we used a simple POF filter to validate the principle of our approach. However, other correlation filters, more efficient, and a large set of subjects will be used i.e. that is the goal of an article in progress [6]. One main drawback of the existing face recognition methods arises from the sensitivity to rotation of the target image with the reference facial images. Several studies have …