Abstract:Multi-spectral face recognition has been attracting increasing interest. In the last decade, several multi-spectral face recognition methods have been presented. However, it has not been well studied that how to jointly learn effective features with favorable discriminability from multiple spectra even when multi-spectral face images are severely contaminated by noise. Multi-view dictionary learning is an effective feature learning technique, which learns dictionaries from multiple views of the same object and has achieved state-of-the-art classification results. In this paper, we for the first time introduce the multi-view dictionary learning technique into the field of multi-spectral face recognition and propose a multi-spectral low-rank structured dictionary learning (MLSDL) approach. It learns multiple structured dictionaries, including a spectrum-common dictionary and multiple spectrum-specific dictionaries, which can fully explore both the correlated information and the complementary information among multiple spectra. Each dictionary contains a set of class-specified sub-dictionaries. Based on the low-rank matrix recovery theory, we apply low-rank regularization in multi-spectral dictionary learning procedure such that MLSDL can well solve the problem of multi-spectral face recognition with high levels of noise. We also design the low-rank structural incoherence term for multi-spectral dictionary learning, so as to reduce the redundancy among multiple spectrum-specific dictionaries. In addition, to enhance the efficiency of classification procedure, we design a low-rank structured collaborative representation classification scheme for MLSDL. Experimental results on HK PolyU, CMU and UWA hyper-spectral face databases demonstrate the effectiveness of the proposed approach.

Face recognition with multi-resolution spectral feature images

Face Recognition with Multi-Resolution Spectral Feature Images.

A spectral feature based approach for face recognition with one training sample

A Local Spectral Feature Based Face Recognition Approach for the One-Sample-per-person Problem.

Sampled Two-Dimensional LDA for Face Recognition with One Training Image per Person

Face Recognition Using a Novel Image Representation Scheme and Multi-Scale Local Features

Face Recognition Using Novel LDA-Based Algorithms.

Robust 3D Face Recognition by Local Shape Difference Boosting

Low-resolution Face Recognition with Single Sample Per Person.

Ensemble of Randomized Linear Discriminant Analysis for Face Recognition with Single Sample Per Person

A New Face Recognition Framework: Symmetrical Bilateral 2DPLS Plus LDA.

Unifying Domain Adaptation into Simultaneous Discriminant Analysis: A Method for Low-Resolution Face Recognition with Single Sample Per Person

Face Recognition Using FLDA with Single Training Image Per Person.

Face Recognition Method Based on Multi-Scale Adaptive LDA

Joint multi-subspace feature learning with singular value decomposition for robust single-sample face recognition

Low-Resolution Face Recognition with Single Sample Per Person Via Domain Adaptation.

Discriminative Multidimensional Scaling for Low-Resolution Face Recognition.

(2D)2LDALPP: A Novel Approach to Face Recognition

Multi-spectral Low-Rank Structured Dictionary Learning for Face Recognition.

Fusing Hierarchical Multi-Scale Local Binary Patterns and Virtual Mirror Samples to Perform Face Recognition