Image face recognition using Hybrid Multiclass SVM (HM-SVM)

Face recognition was one of the most popular topics in the image and video processing research fields. It became main attraction by many researcher due to demand from commercial and law enforcement sectors. The main issue in face recognition are application sensitivity toward intrinsic factors and extrinsic factors. Beside, computation time and memory usage are the important aspect to been consider. This research, introduced a hybrid face recognition technique, which consist of feature extraction and Multiclass Support Vector Machine (M-SVM) classifier. In the first part, Principal Component Analysis (PCA) was used for image dimension reduction and feature extraction. Then two Multiclass Support Vector Machine (M-SVM) strategies were utilized to tackle the face recognition problem. Cambridge ORL Face Database was used which consist of 400 images of 40 individuals. The accuracy evaluation of this research was based on two different SVM kernel types. Comparison was made to classic one-versus-one and bottom-up decision tree Multiclass Support Vector Machine. As a result, the proposed algorithm shown a consistent and higher accuracy rather than classic strategies approximately by 4.5%–18.1% using polynomial kernel and 4.4–20.8% by using radial basis function kernel.