Face Recognition Method Based On SVM And ELM

As a biometric identification technology, face recognition research is animportant branch with challenging, with its intuitive, non-contact and reliabilityare successfully applied in many areas. Face recognition system consists of fourparts: the human face image acquisition, image preprocessing, face imagedetection, face feature extraction, matching and recognition. Advantages anddisadvantages of face recognition algorithm, to a large extent depends on theeffect of human face feature extraction and selection of classifier.This paper focuses on the research of face recognition algorithm of featureextraction phase and classification phase, A system is achieved by using thealgorithm of face recognition, which combines with PCA (Principal ComponentAnalysis) and FastICA (Fixed-Point Independent Component Analysis) toextract face features, and improved SVM (Support Vector Machine) and ELM(Extreme learning machine) is used as classifier. The system of face recognitionused the improved algorithm is tested in the MATLAB2012a software with theORL training and testing database, from two aspects of operation time andrecognition rate to verify the effectiveness of the improved algorithm.This paper contains the main contents as follows:1. In order to eliminate the influence of illumination on the classification performance, translate the images from the three dimensional space into twodimensional space, face detection method based on skin color is used to detectand position the face region in the real-time and high accuracy.2. In the face feature extraction phase, this paper combines principlecomponent analysis and fast independent component analysis to extract humanface features. The first-order discrete wavelet decomposition is adopted toextract low frequency chart of high dimensional face image, the maincharacteristics of information is achieved by using the PCA algorithm, and theindependent face space is obtained by combining the FastICA algorithm, thelocal information of face images are well presented.3. In the classification phase, this paper focuses on two kinds of classifier,support vector machine (SVM) and extreme learning machine (ELM). In facerecognition system based on SVM, aiming at the problem of particle swarmoptimization algorithm is easy to fall into local minimum, introducing thesupport vector machine improved by Geese particle swarm algorithm, theexperimental results show that compared with the traditional algorithmGSPSO-SVM algorithm improves the classification accuracy and reduce thecomputational complexity. In face recognition system based on ELM, in order toovercome the poor robustness of extreme learning machine, this paper putforward the algorithm of face recognition based on SPSO-ELM, theexperimental results prove that the improved algorithm can improve therecognition rate, has good robustness, and effectively improve the classification performance of the extreme learning machine.