Multi-feature characterization strategy for face recognition efficiency

Face recognition became a daily discipline in human life. At the work, with our PDP and smart phones, for our daily help, our security and many other utilities, face recognition has crossed the laboratory doors and colonized the human quotidian. However, the effectiveness of the developed applications still encounters many challenges. The presented work in this paper tries to deal with these challenges by proposing an advanced characterization way to enrich the feature vectors used by the classifiers to verify or identify faces. This process was done by compiling three types of feature vectors. Each type encapsulates a specific type of face information. At first, we compile a feature vector related to the geometric information of the face using Zernike moments; then spectral components using DCT are extracted to form the second type of feature vectors and finally, the last feature vector type is formed by compiling the texture and luminance information using LBP. The three vector types are then combined to form an enriched feature vector which was post-processed through a feature selection method then presented to the input of a neural network classifier. The validation experiments were realized on the XM2VTS and ORL database and recognition rates of 93.3% and 92.5% were respectively recorded for XM2VTS and ORL database.