Coupled Discriminative Feature Learning for Heterogeneous Face Recognition

Jin, Yi; Lu, Jiwen; Ruan, Qiuqi

doi:10.1109/tifs.2015.2390414

Cited by 94 publications

(53 citation statements)

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Besides, the proposed methods are also compared with three other methods [14], [15], [46] that are not based on subspace or metric learning. The method proposed in [46] is image synthesis-based.…”

Section: F Results On Casia Nir-vis 20 Datasetmentioning

confidence: 99%

“…The other two methods are feature learning-based methods. The method in [14] achieved the rank-1 recognition rate of 71.5 ± 1.4% and in [15], the rank-1 recognition rate of 86.2 ± 1.0% was reported. While our results, 73.9 ± 0.9% and 76.0 ± 0.7% were better than that of [14] but worse than that of [15].…”

Section: F Results On Casia Nir-vis 20 Datasetmentioning

confidence: 99%

“…The method in [14] achieved the rank-1 recognition rate of 71.5 ± 1.4% and in [15], the rank-1 recognition rate of 86.2 ± 1.0% was reported. While our results, 73.9 ± 0.9% and 76.0 ± 0.7% were better than that of [14] but worse than that of [15]. Despite of this, as the proposed methods are general metric learning methods, they can be combined with image synthesis-based methods and feature learning methods to take advantages of both.…”

Section: F Results On Casia Nir-vis 20 Datasetmentioning

confidence: 99%

“…Despite of this, as the proposed methods are general metric learning methods, they can be combined with image synthesis-based methods and feature learning methods to take advantages of both. Due to that three methods [14], [15], [46] are not open sources, in the next subsection, we tested the proposed methods when combined with two publicly available deep features. On the CASIA NIR-VIS 2.0 dataset, KMCM 2 L achieved a significant improvement on the rank-1 result of 96.5 ± 0.4% (see Section VII-G and Table VI for details).…”

Section: F Results On Casia Nir-vis 20 Datasetmentioning

confidence: 99%

“…Zhang et al [13] proposed a coupled information-theoretic encoding method to maximize the mutual information between two modalities in the quantized feature spaces. A coupled discriminative feature learning (CDFL) method is proposed in [14]. It learns a few image filters to maximize interclass variations and minimize intraclass variations of the learned feature in a new feature space.…”

Section: ) Modality Invariant Feature Extraction-based Methodsmentioning

confidence: 99%

See 4 more Smart Citations

Heterogeneous Face Recognition by Margin-Based Cross-Modality Metric Learning

Huo

Shi

Yang

et al. 2018

IEEE Trans. Cybern.

View full text Add to dashboard Cite

Abstract-Heterogeneous face recognition deals with matching face images from different modalities or sources. The main challenge lies in cross-modal differences and variations and the goal is to make cross-modality separation among subjects. A margin-based cross-modality metric learning (MCM 2 L) method is proposed to address the problem. A cross-modality metric is defined in a common subspace where samples of two different modalities are mapped and measured. The objective is to learn such metrics that satisfy the following two constraints. The first minimizes pairwise, intrapersonal cross-modality distances. The second forces a margin between subject specific intrapersonal and interpersonal cross-modality distances. This is achieved by defining a hinge loss on triplet-based distance constraints for efficient optimization. It allows the proposed method to focus more on optimizing distances of those subjects whose intrapersonal and interpersonal distances are hard to separate. The proposed method is further extended to a kernelized MCM 2 L (KMCM 2 L). Both methods have been evaluated on an ID card face dataset and two other cross-modality benchmark datasets. Various feature extraction methods have also been incorporated in the study, including recent deep learned features. In extensive experiments and comparisons with the state-of-the-art methods, the MCM 2 L and KMCM 2 L methods achieved marked improvements in most cases.

show abstract

Section: F Results On Casia Nir-vis 20 Datasetmentioning

confidence: 99%

Section: F Results On Casia Nir-vis 20 Datasetmentioning

confidence: 99%