بهبود مدل تفکیک‌کننده منیفلدهای غیرخطی به‌منظور بازشناسی چهره با یک تصویر از هر فرد

Manifold learning is a dimension reduction method for extracting nonlinear structures of high-dimensional data. Many methods have been introduced for this purpose. Most of these methods usually extract a global manifold for data. However, in many real-world problems, there is not only one global manifold, but also additional information about the objects is shared by a large number of manifolds. In this context, based on previous researches, this paper proposes a nonlinear dimension reduction method based on the deep neural network that extract simultaneously manifolds embedded in data. In nonlinear manifold separator model, unlike unsupervised learning of bottleneck neural network, data labels are indirectly used for manifold learning. Given the deep structure of the model, it has been shown that using pre-training methods can significantly improve its performance moreover, to improve within-manifold discrimination for different classes, its standard functions have been improved. This paper makes use of the model for extracting both expression and identity manifolds for facial images of the CK+ database. In comparing early and improved models, it is shown that the facial expression recognition rate from 24.29% to 75.07% and the face recognition rate by a single image of each person by enriching dataset from 90.62% to 97.07% were improved.