In this study, face recognition system is realized with the aid of optimized pRBFNNs focused on hybrid data fusion & 3D recognition base. The recognition rate of existing face recognition systems is usually affected by the change of image factors such as position, scale, backgrounds and pose. Here the face region information obtained from the detected face region is used to compensate for these defects.
The proposed hybrid data fusion method is designed based on two-dimensional imaging techniques by using AdaBoost and the Active Shape Model(ASM) algorithm. The preprocessing part of hybrid data fusion method is given as follows. First, CCD camera is used to obtain a picture frame directly. By using histogram equalization method, the quality of distorted image is enhanced from the influence of natural and artificial illumination.
Second, AdaBoost algorithm is used for the detection of face image between face and non-face image area, and then the contour line and shape of face are extracted by handling ASM. Third, PCA and LDA fusion algorithms are handled to reduce the face image information of high-dimension to that of low-dimension.
In addition, research concerning face image acquisition, preprocessing, feature extraction and recognition was carried out by using 3D scanner to overcome the limitation of face recognition based on two-dimensional image. The preprocessing part of 3D face recognition is given as follows. Facial shape is obtained by using 3D scanner and the pose of distorted facial shape is compensated to the front shape and then depth information of face is extracted through semicircle-based multi-point signature method.
In the recognition part, the Polynomial-based Radial Basis Function Neural Networks(pRBFNNs) is designed to carry out face recognition. The structure of the proposed pRBFNNs consists of three modules of condition, conclusion, and inference phases. The input space of the condition phase is divided by using Fuzzy C-means(FCM) and the local area of the conclusion phase is represented as four types of polynomial functions. The coefficients of connection weights are estimated by Least Square Estimation(LSE) method. The final output of inference phase is obtained through fuzzy inference method. The performance of the proposed RBFNNs is improved by generating nonlinear discriminant function in the output space due to the fuzzy inference mechanism of polynomial-based structure. The essential parameters of the proposed model such as polynomial order type, the number of rules, and fuzzification coefficient are optimized by means of Differential Evolution(DE) as well as Particle Swarm Optimization(PSO).
Real-time face recognition system is designed and the performance of the proposed hybrid data fusion-based face recognition algorithm is evaluated by using yale dataset and IC&CI Lab dataset. Also, the performance of 3D face recognition algorithm is evaluated with the aid of IC&CI Lab dataset.