Unsupervised multi-manifold linear differential projection(UMLDP) for face recognition

A novel efficient algorithm called unsupervised multi-manifold linear differential projection(UMLDP) is proposed to overcome the drawbacks of existing unsupervised linear differential projection(ULDP) for face recognition. Firstly, the multi-manifold local neighborhood graph and the largest global variance is constructed respectively. Next, we calculate a low dimensional manifold embedded in high-dimensional space through the multi-objective optimization. This mapping can not only get the low-dimensional manifolds embedded in a high-dimensional space but also maintain the local and the global structural information effectively. Finally, experimental results validate the effectiveness of the proposed algorithm on the ORL, Yale and AR face databases.