An Effective Face Recognition Framework With Subspace Learning Based on Local Texture Patterns

Currently, there are mainly two kinds of methods for face recognition, i.e. subspace learning methods and local texture methods. Generally, the first are sensitive to compound changes including illumination, face expression and pose, but have low-dimensional features. Relatively, the second have high-dimensional features but better robustness to compound changes. Making full use of their advantages, this paper proposes a face recognition framework with subspace learning based on local texture patterns. Then, two face recognition methods, which adopt discriminant analysis from two different local texture patterns, are derived from the proposed framework. Lastly, the experimental results on AR database and CAS-PEAL-R1 database show that two proposed face recognition methods gain higher average optimal correct recognition rates than several traditional methods, and much lower dimensional features than traditional local texture methods, which verify the effectiveness of two methods derived from the proposed framework. Therefore, the proposed framework is effective and worth extending.