A Face Recognition Algorithm Based on Optimal Feature Selection

Received: 10 January 2019 Accepted: 6 March 2019 To achieve accurate and robust face recognition, this paper designs a face recognition algorithm based on optimal feature selection. The algorithm is denoted as GRA-LSSVM, because it integrates grey relational analysis (GRA) with least squares support vector machine (LSSVM). Firstly, the target face image was segmented into several subblocks. Next, the global features of the face were extracted from each subblock by kernel principal component analysis (PCA). After that, the GRA algorithm was introduced to determine the features that contribute greatly to face recognition. These features were integrated into an eigenvector. Finally, a face classifier by the LSSVM based on the “one-to-many” principle, and simulated with multiple face databases. The simulation shows the GRA-LSSVM derived the optimal feature subset for face recognition, and thus outperformed other face recognition algorithms in accuracy and speed. The research provides an effective and advanced method for face recognition.