The face image modeling by eigenvalues is not a new track in the literature. However, a much complete study is required to achieve a comprehensive investigation of the topic. In this research paper, an experimental methodology is conducted for studying the different alternatives of utilizing the eigenvalues for human face recognition. For a better universal investigation, three popular databases are tested; Orl_faces, extended Yale face_A, and extended Yale face_B datasets. The main objective of the study is to find the best choice of using eigenvalues (EV) in face recognition. The technique of the moving average filter (MAF) is combined with that of eigenvalues to enhance the results. Probabilistic neural network (PNN) is used for classification. Three methods of this concept were developed as follows: EV, EV with MAF, and MAF alone. The elapsed time was tested, where for moving average filter was distinctly smaller than the other two methods. For the Yaleface_B database, the eigenvalues method was superior for each of the three training/testing systems. The results were enhanced after using different filters instead of a direct moving average filter to make the proposed method the superior again. The study proved the possibility of using eigenvalues in conjunction with a suitable filter to get acceptable results for all types of image limitations. The concluded ideas elicited from the study spot the light on the usefulness of utilization of eigenvalues in the face recognition tasks.
A Practical Approach for Identity-Embodied 3D Artistic Face Modeling