1 October 2019 in International Journal of Machine Learning and Computing
15 October 2019 in IEEE Access
Set-based image classification has become popular in recent years since it can provide a relatively large amount of within-set information that benefits classification. Grassmann Discriminant Analysis (GDA) models image sets as points (subspaces) on a Grassmann manifold and then explicitly maps them to a higher-dimensional Hilbert space, where Euclidean geometry applies, for Linear Discriminant Analysis (LDA). However, due to the noise disturbance and finite number of training sammples in practice, the conventional problems of LDA, such as the singularity of the within-class scatter matrix and the instability of its inverse, also appear with GDA, which result in recognition performance deterioration. Inspired by eigenspectrum regularization techniques, we propose an eigenspectrum Regularized GDA (RGDA) method to alleviate the conventional problems of GDA in Grassmannian space. Moreover, we implement it with the graph embedded framework, three different eigenspectrum regularization models are incorporated into the proposed approach respectively. Extensive experimental results on set-based face recognition and object categorization tasks have confirmed the effectiveness of our approaches.
29 May 2018 in IEEE Access
Great success in face recognition has been achieved in recent years; however, complex variations and low-resolution images remain a challenge for unconstrained face recognition. Face recognition in video or image sets, which is known as image-set-based face recognition (ISFR), is one feasible solution to address this problem. Regularized nearest points (RNP) is an effective hull-based ISFR method which uses linear space as the input. However, nonlinearity usually exists when the input data contain complex structures, such as illumination and pose variations. Hence, we propose to map the input data to a higher-dimensional feature space by using kernel functions, and we develop the kernel extension of the efficient iterative solver to find the regularized nearest points between two sets in higher-dimensional feature space. We also exploit this kernel efficient iterative solver to improve the kernel convex hull image-set-based collaborative representation and classification (KCH-ISCRC) method. The proposed kernelized fast algorithm improves the face recognition ability of RNP and significantly accelerates the kernel version hull-based ISFR methods. Experiments are performed on three benchmark face recognition video datasets. The experimental results illustrate the effectiveness of our proposed methods.