Heterogeneous Face Research Articles

Homomorphic Filtering and Phase-Based Matching for Cross-Spectral Cross-Distance Face Recognition.

Facial recognition has a significant application for security, especially in surveillance technologies. In surveillance systems, recognizing faces captured far away from the camera under various lighting conditions, such as in the daytime and nighttime, is a challenging task. A system capable of recognizing face images in both daytime and nighttime and at various distances is called Cross-Spectral Cross Distance (CSCD) face recognition. In this paper, we proposed a phase-based CSCD face recognition approach. We employed Homomorphic filtering as photometric normalization and Band Limited Phase Only Correlation (BLPOC) for image matching. Different from the state-of-the-art methods, we directly utilized the phase component from an image, without the need for a feature extraction process. The experiment was conducted using the Long-Distance Heterogeneous Face Database (LDHF-DB). The proposed method was evaluated in three scenarios: (i) cross-spectral face verification at 1m, (ii) cross-spectral face verification at 60m, and (iii) cross-spectral face verification where the probe images (near-infrared (NIR) face images) were captured at 1m and the gallery data (face images) was captured at 60 m. The proposed CSCD method resulted in the best recognition performance among the CSCD baseline approaches, with an Equal Error Rate (EER) of 5.34% and a Genuine Acceptance Rate (GAR) of 93%.

Basket of options: Unpacking the concept.

How to stimulate technological change to enhance agricultural productivity and reduce poverty remains an area of vigorous debate. In the face of heterogeneity among farm households and rural areas, one proposition is to offer potential users a ‘basket of options’ – a range of agricultural technologies from which potential users may select the ones that are best suited to their specific circumstances. While the idea of a basket of options is now generally accepted, it has attracted little critical attention. In this paper, we reflect on outstanding questions: the appropriate dimensions of a basket, its contents and how they are identified, and how a basket might be presented. We conceive a basket of options in terms of its depth (number of options related to a problem or opportunity) and breadth (the number of different problems or opportunities addressed). The dimensions of a basket should reflect the framing of the problem or opportunity at hand and the objective in offering the basket. We recognise that increasing the number of options leads to a trade-off by decreasing the fraction of those options that are relevant to an individual user. Farmers might try out, adapt or use one or more of the options in a basket, possibly leading to a process of technological change. We emphasise that the selection (or not) of specific options from the basket, and potential adaptation of the options, provide important opportunities for learning. Baskets of options can therefore be understood as important boundary concepts that invite critical engagement, comparison and discussion. Significant knowledge gaps remain, however, about the best ways to present the basket and to guide potential users to select the options that are most relevant to them.

Finding Treatment Effects in Alzheimer Trials in the Face of Disease Progression Heterogeneity.

ObjectiveTo investigate the influence of heterogeneity in disease progression for detecting treatment effects in Alzheimer disease (AD) trials, using a simulation study.MethodsIndividuals with an abnormal amyloid PET scan, diagnosis of mild cognitive impairment or dementia, baseline Mini-Mental State Examination (MMSE) score ≥24, global Clinical Dementia Rating (CDR) score of 0.5, and ≥1 follow-up cognitive assessment were selected from the Alzheimer's Disease Neuroimaging Initiative database (n = 302, age 73 ± 6.7; 44% female; 16.1 ± 2.7 years of education; 69% APOE ε4 carrier). We simulated a clinical trial by randomly assigning individuals to a “placebo” and “treatment” group and subsequently computed group differences on the CDR–sum of boxes (CDR-SB), Alzheimer’s Disease Assessment Scale–cognitive subscale–13 and MMSE after 18 months follow-up. We repeated this simulation 10,000 times to determine the 95% range of effect sizes. We further studied the influence of known AD risk factors (age, sex, education, APOE ε4 status, CSF total tau levels) on the variability in effect sizes.ResultsIndividual trajectories on all cognitive outcomes were highly variable, and the 95% ranges of possible effect sizes at 18 months were broad (e.g., ranging from 0.35 improvement to 0.35 decline on the CDR-SB). Results of recent anti-amyloid trials mostly fell within these 95% ranges of effect sizes. APOE ε4 carriers and individuals with abnormal baseline tau levels showed faster decline at group level, but also greater within-group variability, as illustrated by broader 95% effect size ranges (e.g., ±0.70 points for the CDR-SB).ConclusionsIndividuals with early AD show heterogeneity in disease progression, which increases when stratifying on risk factors associated with progression. We provide guidance for a priori effect sizes on cognitive outcomes for detecting true change, which is crucial for future AD trials.

Dual Face Alignment Learning Network for NIR-VIS Face Recognition

As the most important topic in Heterogeneous Face Recognition (HFR), Near-InfraRed and VISual (NIR-VIS) face recognition has attracted increasing research attentions owing to its potential application in the field of criminal detective cases and multimedia information retrieval. However, due to its dramatic intra-class variations, including modality, pose, occlusion, blurry, lighting, distance, expression, etc, it is very challenging to retain inherent identity information. To address the above issue, we propose a novel Dual Face Alignment Learning (DFAL) algorithm to explore the potential domain-invariant neutral face representations of the cross-modal images. Our model contains three effective components including Feature-level Face Alignment (FFA), Image-level Face Alignment (IFA) and Cross-domain compact Representation (CdR). Firstly, Teacher-Encoder CNNs (TeEn-CNNs) and Student-Encoder CNNs (StEn-CNNs) are designed to encode features for VIS neutral face images and non-neutral face images, and the FFA is introduced to learn neutral face representations by performing feature-level alignment between non-neutral face and VIS neutral face. Secondly, Student-Decoder CNNs (StDe-CNNs) is developed to decode features to restore face images, and the IFA is designed to reconstruct neutral face image by imposing image-level alignment. Notably, the FFA acts as the primary target to learn VIS neutral face representations for cross-view data, while the IFA plays a role in the icing on the cake, i.e., further disentangling domain and residual information through the synthesis process. Finally, the CdR dispels modality features and distills identity features by mining inter-class information, inter-domain information and inter-semantic relationship. The joint scheme enables the elimination of intra-class variations and the purification of identity information. We carry out comprehensive experiments to illustrate the effectiveness of the DFAL approach on three challenging NIR-VIS databases.

Developing hierarchical density‐structured models to study the national‐scale dynamics of an arable weed

AbstractPopulation dynamics can be highly variable in the face of environmental heterogeneity, and understanding this variation is central in the study of ecology. Robust management decisions require that we understand how populations respond to management at a range of scales, and under a broad suite of conditions. Population models are potentially valuable tools in addressing this challenge. However, without adequate data, models can fail to produce useful results. Populations of arable weeds are particularly problematic in this respect, as they are widespread and their dynamics are extremely variable. Owing to the inherent cost of collecting data, most studies of plant population dynamics are derived from localized experiments under a small range of environmental conditions, limiting the extent to which variance in population dynamics can be measured. Density‐structured models provide a route to rapid, large‐scale analysis of population dynamics, and can expand the scale of ecological models that are directly tied to data. Here we extend previous density‐structured models to include environmental heterogeneity, variation in management, and to account for inter‐population variation. We develop, parameterize, and test hierarchical density‐structured models for a common agricultural weed, black‐grass (Alopecurus myosuroides). We model the dynamics of this species in response to crop management, using survey data gathered over 4 yr from 364 fields across a network of 45 UK farms. We show that hierarchical density‐structured models provide a substantial improvement over their nonhierarchical counterparts. Using these models, we demonstrate that several alternative crop rotations are effective in reducing weed densities. Rotations with high wheat prevalence exhibit the most severe infestations, and diverse rotations generally have lower weed densities. However, a key outcome is that in many cases the effect of crop rotation is small compared to the high variability arising from spatiotemporal heterogeneity. This result highlights the need to monitor and model population dynamics across large spatial and temporal scales in order to account for variation in the drivers of plant dynamics. Our framework for data collection and modeling provides a means to achieve this.

Background complexity and optimal background matching camouflage

Camouflage through background matching is a widespread antipredator strategy in which animals blend in with their background to avoid detection. To maximise survival in a variable natural environment, animals can have colourations that either match one of the backgrounds maximally (i.e. specialist strategy) or match multiple backgrounds partially (i.e. generalist strategy). Theoretical work indicates that the optimal strategy depends on the extent of visual difference between the backgrounds (i.e. heterogeneity) or how commonly the animal will encounter the background types. However, the role of another critical determinant of detection, the visual complexity of the background, on optimal camouflage strategy (specialist versus generalist) in the face of background heterogeneity, remains unknown. Here, we performed a virtual predation experiment employing humans as surrogate ‘predators’ and explored how background complexity influences camouflage in heterogeneous backgrounds. Under low heterogeneity, we found the latency to attack generalists was higher than that for specialists on a complex background, but there was no difference between specialists and generalists on a simple background. At intermediate heterogeneity, both specialist and generalist targets took a similar time to be attacked irrespective of complexity, suggesting that both the strategies may co-exist. In contrast, at high levels of heterogeneity, we found generalists were attacked sooner when compared to specialists irrespective of whether the background was simple or complex. Our results thus suggest that complex backgrounds favour the evolution of a generalist background matching strategy that maximises fitness in multiple backgrounds but only when the visual difference between the backgrounds is low. Overall, our study provides key insights highlighting the underappreciated role of background complexity on the optimization and evolution of camouflage colouration in a heterogeneous environment. Many animals often face the challenge of encountering multiple visually distinct backgrounds due to variation in their environment, i.e. background heterogeneity. How should animals optimise camouflage when there is background heterogeneity? Theoretical studies have proposed that animals may match one of the many backgrounds (specialise) or match multiple backgrounds partially (generalise) as an optimal solution. However, cognitive constraints from the predator’s perspective may also have a role to play in this optimization problem, but this has not been examined. Our experiments involving humans as ‘predators’ show that when background complexity renders the search task more difficult, generalist targets took a longer time to be attacked than specialist targets, but only in less heterogeneous backgrounds. However, irrespective of complexity, specialist targets are better than generalists at avoiding attack in highly heterogeneous backgrounds. Cognitive constraints of predators may, therefore, play a significant role in the optimization of camouflage colouration in heterogeneous environments.

Benchmarking lightweight face architectures on specific face recognition scenarios

This paper studies the impact of lightweight face models on real applications. Lightweight architectures proposed for face recognition are analyzed and evaluated on different scenarios. In particular, we evaluate the performance of five recent lightweight architectures on five face recognition scenarios: image and video based face recognition, cross-factor and heterogeneous face recognition, as well as active authentication on mobile devices. In addition, we show the lacks of using common lightweight models unchanged for specific face recognition tasks, by assessing the performance of the original lightweight versions of the lightweight face models considered in our study. We also show that the inference time on different devices and the computational requirements of the lightweight architectures allows their use on real-time applications or computationally limited platforms. In summary, this paper can serve as a baseline in order to select lightweight face architectures depending on the practical application at hand. Besides, it provides some insights about the remaining challenges and possible future research topics.

LAMP-HQ: A Large-Scale Multi-pose High-Quality Database and Benchmark for NIR-VIS Face Recognition

Near-infrared-visible (NIR-VIS) heterogeneous face recognition matches NIR to corresponding VIS face images. However, due to the sensing gap, NIR images often lose some identity information so that the NIR-VIS recognition issue is more difficult than conventional VIS face recognition. Recently, NIR-VIS heterogeneous face recognition has attracted considerable attention in the computer vision community because of its convenience and adaptability in practical applications. Various deep learning-based methods have been proposed and substantially increased the recognition performance, but the lack of NIR-VIS training samples leads to the difficulty of the model training process. In this paper, we propose a new $$\mathbf{L} {} \mathbf{a} $$ rge-Scale $$\mathbf{M} $$ ulti- $$\mathbf{P} $$ ose $$\mathbf{H} $$ igh- $$\mathbf{Q} $$ uality NIR-VIS database ‘ $$\mathbf{LAMP}-HQ $$ ’ containing 56,788 NIR and 16,828 VIS images of 573 subjects with large diversities in pose, illumination, attribute, scene and accessory. We furnish a benchmark along with the protocol for NIR-VIS face recognition via generation on LAMP-HQ, including Pixel2-Pixel, CycleGAN, ADFL, PCFH, and PACH. Furthermore, we propose a novel exemplar-based variational spectral attention network to produce high-fidelity VIS images from NIR data. A spectral conditional attention module is introduced to reduce the domain gap between NIR and VIS data and then improve the performance of NIR-VIS heterogeneous face recognition on various databases including the LAMP-HQ.

Gammadion binary pattern of Shearlet coefficients (GBPSC): An illumination-invariant heterogeneous face descriptor

• Usage of Shearlet coefficients is proposed to cancel illumination and noise variations. • Novel Gammadion Binary Pattern is proposed to extract out local micro level features. • GBPSC descriptor is devised to capture modality-invariant patterns of facial features. • GBPSC along with CNN outperforms other competing methods even with a small dataset. This paper presents a novel face image descriptor called Gammadion Binary Pattern of Shearlet Coefficients (GBPSC) for illumination and noise invariant, homogeneous and heterogeneous face recognition. Exploiting the energy concentration property of the Digital Shearlet Transform, an efficient illumination and noise invariant feature extractor has been devised. Finally, inspired by the Gammadion structure, a robust multi-directional local binary pattern named Gammadion Binary Pattern (GBP) has been proposed. GBP is applied on the previously extracted illumination and noise invariant feature map to generate the GBPSC images. Recognition results on Extended Yale B and TUFTS dataset indicate the primacy of the proposed scheme in terms of common feature representation under varying illumination, and modality. Furthermore, the merger of the proposed GBPSC and Convolutional Neural Network (CNN) consistently outperforms other state-of-the art methods.

Surface settlements caused by mechanised tunnelling in hard rock with heterogeneous tunnel face conditions – Empirical and numerical calculations using the example of Boßler Tunnel

AbstractAn empirical and a numerical calculation model were developed to predict the surface settlements due to EPB tunnelling in hard rock with heterogeneous face conditions. For this purpose, surface settlement and TBM driving data from Boßler Tunnel were considered. The TBM launching area characterised by a low overburden (z/D < 2) and a nature conservation area at surface was examined. The developed models were calibrated against the measured surface settlements of the first eastbound tube and were used to predict the surface settlements of the subsequent westbound tube. The prediction was then compared to the actual measured settlements during the excavation of the westbound tube. From this, findings for future settlement predictions in similar ground conditions were derived. It could be shown that both models are suitable for the application in the investigated subsoil conditions and can be used successfully in practice in different phases of the project development.

SADG: Self-Aligned Dual NIR-VIS Generation for Heterogeneous Face Recognition

Heterogeneous face recognition (HFR) has aroused significant interest in recent years, with some challenging tasks such as misalignment problems and limited HFR data. Misalignment occurs among different modalities’ images mainly because of misaligned semantics. Although recent methods have attempted to settle the low-shot problem, they suffer from the misalignment problem between paired near infrared (NIR) and visible (VIS) images. Misalignment can bring performance degradation to most image-to-image translation networks. In this work, we propose a self-aligned dual generation (SADG) architecture for generating semantics-aligned pairwise NIR-VIS images with the same identity, but without the additional guidance of external information learning. Specifically, we propose a self-aligned generator to align the data distributions between two modalities. Then, we present a multiscale patch discriminator to get high quality images. Furthermore, we raise the mean landmark distance (MLD) to test the alignment performance between NIR and VIS images with the same identity. Extensive experiments and an ablation study of SADG on three public datasets show significant alignment performance and recognition results. Specifically, the Rank1 accuracy achieved was close to 99.9% for the CASIA NIR-VIS 2.0, Oulu-CASIA NIR-VIS and BUAA VIS-NIR datasets, respectively.

Revisiting Bertrand Competition in the Face of Customer Heterogeneity and Price Discrimination

This paper compares uniform vs. discriminative pricing strategies in a competitive setting featuring capacity restrictions, quality differentiation, and customer heterogeneity. It tackles this overarching question through operational lenses, by studying the joint impact of price discrimination and the allocation of limited inventory across customer segments. Thanks to advances in information technology, firms have developed new capabilities in terms of customer segmentation (by clustering customers into different segments) and price discrimination (by offering different prices to different customer segments). Price discrimination, although seemingly beneficial, can hurt firms in competitive environments. It is thus critical for firms to understand when to engage in price discrimination, and how to support discriminative pricing practices with appropriate inventory management strategies. We develop a Bertrand competition game featuring capacity restrictions, quality differentiation, and customer heterogeneity. We characterize (pure- or mixed-strategy) Nash equilibria for a single-stage game reflecting uniform pricing and for a two-stage inventory-price game reflecting discriminative pricing. We identify three sources of market friction in price competition, enabling firms to earn higher profits: capacity limitations, quality differentiation, and customer heterogeneity. Price discrimination eliminates the market frictions from customer heterogeneity, but strategic inventory allocation restores (or strengthens) the market frictions from capacity limitations. Ultimately, price discrimination is only beneficial when combined with optimal inventory allocation, thus underscoring the critical role of inventory allocation in the design of competitive pricing strategies.

DVG-Face: Dual Variational Generation for Heterogeneous Face Recognition.

Heterogeneous face recognition (HFR) refers to matching cross-domain faces and plays a crucial role in public security. Nevertheless, HFR is confronted with challenges from large domain discrepancy and insufficient heterogeneous data. In this paper, we formulate HFR as a dual generation problem, and tackle it via a novel dual variational generation (DVG-Face) framework. Specifically, a dual variational generator is elaborately designed to learn the joint distribution of paired heterogeneous images. However, the small-scale paired heterogeneous training data may limit the identity diversity of sampling. In order to break through the limitation, we propose to integrate abundant identity information of large-scale visible data into the joint distribution. Furthermore, a pairwise identity preserving loss is imposed on the generated paired heterogeneous images to ensure their identity consistency. As a consequence, massive new diverse paired heterogeneous images with the same identity can be generated from noises. The identity consistency and identity diversity properties allow us to employ these generated images to train the HFR network via a contrastive learning mechanism, yielding both domain-invariant and discriminative embedding features. Concretely, the generated paired heterogeneous images are regarded as positive pairs, and the images obtained from different samplings are considered as negative pairs. Our method achieves superior performances over state-of-the-art methods on seven challenging databases belonging to five HFR tasks, including NIR-VIS, Sketch-Photo, Profile-Frontal Photo, Thermal-VIS, and ID-Camera.

An Applied Researcher’s Guide to Estimating Effects from Multisite Individually Randomized Trials: Estimands, Estimators, and Estimates

Researchers face many choices when conducting large-scale multisite individually randomized control trials. One of the most common quantities of interest in multisite RCTs is the overall average effect. Even this quantity is non-trivial to define and estimate. The researcher can target the average effect across individuals or sites. Furthermore, the researcher can target the effect for the experimental sample or a larger population. If treatment effects vary across sites, these estimands can differ. Once an estimand is selected, an estimator must be chosen. Standard estimators, such as fixed-effects regression, can be biased. We describe 15 estimators, consider which estimands they are appropriate for, and discuss their properties in the face of cross-site effect heterogeneity. Using data from 12 large multisite RCTs, we estimate the effect (and standard error) using each estimator and compare the results. We assess the extent that these decisions matter in practice and provide guidance for applied researchers.

Learning With Heterogeneous Misspecified Models: Characterization and Robustness

This paper develops a general framework to study how misinterpreting information impacts learning. Our main result is a simple criterion to characterize long‐run beliefs based on the underlying form of misspecification. We present this characterization in the context of social learning, then highlight how it applies to other learning environments, including individual learning. A key contribution is that our characterization applies to settings with model heterogeneity and provides conditions for entrenched disagreement. Our characterization can be used to determine whether a representative agent approach is valid in the face of heterogeneity, study how differing levels of bias or unawareness of others' biases impact learning, and explore whether the impact of a bias is sensitive to parametric specification or the source of information. This unified framework synthesizes insights gleaned from previously studied forms of misspecification and provides novel insights in specific applications, as we demonstrate in settings with partisan bias, overreaction, naive learning, and level‐k reasoning.

Learning with Heterogeneous Misspecified Models: Characterization and Robustness

This paper develops a general framework to study how misinterpreting information impacts learning. Our main result is a simple criterion to characterize long-run beliefs based on the underlying form of misspecification. We present this characterization in the context of social learning, then highlight how it applies to other learning environments, including individual learning. A key contribution is that our characterization applies to settings with model heterogeneity and provides conditions for entrenched disagreement. Our characterization can be used to determine whether a representative agent approach is valid in the face of heterogeneity, study how differing levels of bias or unawareness of others’ biases impact learning, and explore whether the impact of a bias is sensitive to parametric specification or the source of information. This unified framework synthesizes insights gleaned from previously studied forms of misspecification and provides novel insights in specific applications, as we demonstrate in settings with partisan bias, overreaction, naive learning, and level-k reasoning.

Weighted Feature Histogram of Multi-Scale Local Patch Using Multi-Bit Binary Descriptor for Face Recognition.

Most face recognition methods employ single-bit binary descriptors for face representation. The information from these methods is lost in the process of quantization from real-valued descriptors to binary descriptors, which greatly limits their robustness for face recognition. In this study, we propose a novel weighted feature histogram (WFH) method of multi-scale local patches using multi-bit binary descriptors for face recognition. First, to obtain multi-scale information of the face image, the local patches are extracted using a multi-scale local patch generation (MSLPG) method. Second, with the goal of reducing the quantization information loss of binary descriptors, a novel multi-bit local binary descriptor learning (MBLBDL) method is proposed to extract multi-bit local binary descriptors (MBLBDs). In MBLBDL, a learned mapping matrix and novel multi-bit coding rules are employed to project pixel difference vectors (PDVs) into the MBLBDs in each local patch. Finally, a novel robust weight learning (RWL) method is proposed to learn a set of robust weights for each patch to integrate the MBLBDs into the final face representation. In RWL, a codebook is first constructed by clustering MBLBDs on each local patch to extract a feature histogram. Then, considering that different parts of the face have different degrees of robustness to local changes, a set of weights is learned to concatenate the feature histograms of all local patches into the final representation of a face image. In addition, to further improve the performance for heterogeneous face recognition, a coupled WFH (C-WFH) method is proposed. C-WFH maintains the similarity of the corresponding MBLBDs and feature histograms for a pair of heterogeneous face images by means of a novel coupled feature learning (CFL) method to reduce the modality gap. A series of experiments are conducted on widely used face datasets to analyze the performance of WFH and C-WFH. Extensive experimental results show that WFH and C-WFH outperform state-of-the-art face recognition methods.

A Study on the Performance of Unconstrained Very Low Resolution Face Recognition: Analyzing Current Trends and New Research Directions

In the past decade, research in the face recognition area has advanced tremendously, particularly in uncontrolled scenarios (face recognition in the wild). This advancement has been achieved partly due to the massive popularity and effectiveness of deep convolutional neural networks and the availability of larger unconstrained datasets. However, several face recognition challenges remain in the context of very low resolution homogeneous (same domain) and heterogeneous (different domain) face recognition. In this survey, we study the seminal and novel methods to tackle the very low resolution face recognition problem and provide an in-depth analysis of their design, effectiveness, and efficiency for a real-time surveillance application. Furthermore, we analyze the advantage of employing deep learning convolutional neural networks, while presenting future research directions for effective deep learning network design in this context.

Partial NIR-VIS Heterogeneous Face Recognition With Automatic Saliency Search

Near-infrared-visual (NIR-VIS) heterogeneous face recognition (HFR) aims to match NIR face images with the corresponding VIS ones. It is a challenging task due to the sensing gaps among different modalities. Occlusions in the input face images make the task extremely complex. To tackle these problems, we present a Saliency Search Network (SSN) to extract domain-invariant identity features. We propose to automatically search the efficient parts of face images in a modality-aware manner, and remove redundant information. Moreover, the searching process is guided by an information bottleneck network, which mitigates the overfitting problems caused by small datasets. Extensive experiments on both complete and partial NIR-VIS HFR on multiple datasets demonstrate the effectiveness and robustness of the proposed method to modality discrepancy and occlusions.

Towards NIR-VIS Masked Face Recognition

Near-infrared to visible (NIR-VIS) face recognition is the most common case in heterogeneous face recognition, which aims to match a pair of face images captured from two different modalities. Existing deep learning based methods have made remarkable progress in NIR-VIS face recognition, while it encounters certain newly-emerged difficulties during the pandemic of COVID-19, since people are supposed to wear facial masks to cut off the spread of the virus. We define this task as NIR-VIS masked face recognition, and find it problematic with the masked face in the NIR probe image. First, the lack of masked face data is a challenging issue for the network training. Second, most of the facial parts (cheeks, mouth, nose etc.) are fully occluded by the mask, which leads to a large amount of loss of information. Third, the domain gap still exists in the remaining facial parts. In such scenario, the existing methods suffer from significant performance degradation caused by the above issues. In this paper, we aim to address the challenge of NIR-VIS masked face recognition from the perspectives of training data and training method. Specifically, we propose a novel heterogeneous training method to maximize the mutual information shared by the face representation of two domains with the help of semi-siamese networks. In addition, a 3D face reconstruction based approach is employed to synthesize masked face from the existing NIR image. Resorting to these practices, our solution provides the domain-invariant face representation which is also robust to the mask occlusion. Extensive experiments on three NIR-VIS face datasets demonstrate the effectiveness and cross-dataset-generalization capacity of our method.