Face Recognition Based on Randomized Subspace Feature

Kernel Principal Component Analysis (KPCA) is a popular feature extraction technique for face recognition. However, it often suffers from the high computational complexity problem, when dealing with large samples. Besides, KPCA is a holistic feature based approach, which means that it discards some useful discriminate local information. In this paper, we use Random Nonlinear Principal Component Analysis (RNPCA) and extract Local Ternary Patterns (LTP) features to improve them respectively. We calculate the kernel matrix by constructing random Fourier features, thus the computation efficiency is speeded up. The LTP features are also extracted, so the local texture information is preserved. In the classification section, we use distance metric learning to improve the classification ability of nearest neighbors classifier. Experimental results on AR, FERET, Yale, ORL face databases demonstrated the effectiveness of our method.