Abstract:Practical video scene and face recognition systems are sometimes confronted with low-resolution (LR) images. The faces may be very small even if the video is clear, thus it is difficult to directly measure the similarity between the faces and the high-resolution (HR) training samples. Traditional super-resolution (SR) methods based face recognition usually have limited performance because the target of SR may not be consistent with that of classification, and time-consuming SR algorithms are not suitable for real-time applications. In this paper, a new feature extraction method called Coupled Kernel Embedding (CKE) is proposed for LR face recognition without any SR preprocessing. In this method, the final kernel matrix is constructed by concatenating two individual kernel matrices in the diagonal direction, and the (semi-)positively definite properties are preserved for optimization. CKE addresses the problem of comparing multi-modal data that are difficult for conventional methods in practice due to the lack of an efficient similarity measure. Particularly, different kernel types (e.g., linear, Gaussian, polynomial) can be integrated into an uniformed optimization objective, which cannot be achieved by simple linear methods. CKE solves this problem by minimizing the dissimilarities captured by their kernel Gram matrices in the low- and high-resolution spaces. In the implementation, the nonlinear objective function is minimized by a generalized eigenvalue decomposition. Experiments on benchmark and real databases show that our CKE method indeed improves the recognition performance.

Lower Bounds Estimation to KL Transform in Face Representation and Recognition

Human Face Recognition Based on Karhunen-Loeve Transform

Study on the Performance of Moments As Invariant Descriptors for Practical Face Recognition Systems

Kernel Low-Rank Representation for Face Recognition.

Constrained Linear Discrimination Analysis and Face Recognition

Face recognition using a kernel fractional-step discriminant analysis algorithm

Face Recognition Using Gabor-based Complete Kernel Fisher Discriminant Analysis with Fractional Power Polynomial Models

Kernel linear regression for face recognition

Research of automated face recognition based on K-L transform

Improving Karhunen -Loeve Transform of Images Based on Neural Networks

Low Computation Face Verification Using Class Center Analysis

Coupled kernel embedding for low resolution face image recognition

Sequential Updating Algorithm for Extracting the Basis of Karhunen Loeve Transformation

Face Recognition Technology Used in Digital Video Record System

Nonlinear Low Rank Representation by Estimating Optimal Transformations for Image Clustering

Laplacian Smoothing Transform for Face Recognition

Domain-invariant Face Recognition using Learned Low-rank Transformation

Fast <formula Formulatype="inline"><tex Notation="TeX">$\ell_{1}$</tex></formula>-Minimization Algorithms for Robust Face Recognition

Heat Kernel Based Local Binary Pattern for Face Representation

Local Gabor Binary Patterns Based on Mutual Information for Face Recognition

A Face Recognition approach based on entropy estimate of the nonlinear DCT features in the Logarithm Domain together with Kernel Entropy Component Analysis