Target Face Research Articles

Enabling witnesses to actively explore faces and reinstate study-test pose during a lineup increases discriminability

Accurate witness identification is a cornerstone of police inquiries and national security investigations. However, witnesses can make errors. We experimentally tested whether an interactive lineup, a recently introduced procedure that enables witnesses to dynamically view and explore faces from different angles, improves the rate at which witnesses identify guilty over innocent suspects compared to procedures traditionally used by law enforcement. Participants encoded 12 target faces, either from the front or in profile view, and then attempted to identify the targets from 12 lineups, half of which were target present and the other half target absent. Participants were randomly assigned to a lineup condition: simultaneous interactive, simultaneous photo, or sequential video. In the front-encoding and profile-encoding conditions, Receiver Operating Characteristics analysis indicated that discriminability was higher in interactive compared to both photo and video lineups, demonstrating the benefit of actively exploring the lineup members' faces. Signal-detection modeling suggested interactive lineups increase discriminability because they afford the witness the opportunity to view more diagnostic features such that the nondiagnostic features play a proportionally lesser role. These findings suggest that eyewitness errors can be reduced using interactive lineups because they create retrieval conditions that enable witnesses to actively explore faces and more effectively sample features.

ONE SHOT FACE STYLIZATION USING GANS

One-shot face stylization is an interesting and challenging subject in computer vision and deep learning. This work deals with the art of manipulating a target face using a reference image as inspiration, which requires controlling facial recognition while specifying important style characteristics This project has attracted a lot of interest due to its potential applications in digital art, entertainment, and personal products. In this abstract, we examine the important features of a one- shot face stylization. Deep neural networks, especially generative adversarial networks (GANs), are widely used in the process to generate customized facial images. These networks are trained on data structures that combine the target and reference faces, with the reference image acting as a strategic identifier. The success of a one-shot facial lies in the meticulous execution of the fading process, which strikes a balance between preserving identity and improving technique. These disadvantages typically include a combination of manpower retention, strategic formation, emotional quality, and enemy training. In conclusion, advances in this field have the potential to transform creative expression and personalization across industries from digital art and animation to virtual avatars and social media filters. Key Words: Facial recognition, generative adversarial networks, virtual avatar, image to image transfer.

Examining StyleGAN as a Utility-Preserving Face De-identification Method

Several face de-identification methods have been proposed to preserve users’ privacy by obscuring their faces. These methods, however, can degrade the quality of photos, and they usually do not preserve the utility of faces, i.e., their age, gender, pose, and facial expression. Recently, advanced generative adversarial network models, such as StyleGAN [ 33], have been proposed, which generate realistic, high-quality imaginary faces. In this paper, we investigate the use of StyleGAN in generating de-identified faces through style mixing, where the styles or features of the target face and an auxiliary face get mixed to generate a de-identified face that carries the utilities of the target face. We examined this de-identification method for preserving utility and privacy by implementing several face detection, verification, and identification attacks and conducting a user study. The results from our extensive experiments, human evaluation, and comparison with two state-of-the-art face de-identification methods, i.e., CIAGAN and DeepPrivacy, show that StyleGAN performs on par or better than these methods, preserving users’ privacy and images’ utility. In particular, the results of the machine learning-based experiments show that StyleGAN0-4 preserves utility better than CIAGAN and DeepPrivacy while preserving privacy at the same level. StyleGAN 0-3 preserves utility at the same level while providing more privacy. In this paper, for the first time, we also performed a carefully designed user study to examine both privacy and utility-preserving properties of StyleGAN 0-3, 0-4, and 0-5, as well as CIAGAN and DeepPrivacy from the human observers’ perspectives. Our statistical tests showed that participants tend to verify and identify StyleGAN 0-5 images easier than DeepPrivacy images. All the methods but StyleGAN 0-5 had significantly lower identification rates than CIAGAN. Regarding utility, as expected, StyleGAN 0-5 performed significantly better in preserving some attributes. Among all methods, on average, participants believe gender has been preserved the most while naturalness has been preserved the least.

Semantics‐guided generative diffusion model with a 3DMM model condition for face swapping

AbstractFace swapping is a technique that replaces a face in a target media with another face of a different identity from a source face image. Currently, research on the effective utilisation of prior knowledge and semantic guidance for photo‐realistic face swapping remains limited, despite the impressive synthesis quality achieved by recent generative models. In this paper, we propose a novel conditional Denoising Diffusion Probabilistic Model (DDPM) enforced by a two‐level face prior guidance. Specifically, it includes (i) an image‐level condition generated by a 3D Morphable Model (3DMM), and (ii) a high‐semantic level guidance driven by information extracted from several pre‐trained attribute classifiers, for high‐quality face image synthesis. Although swapped face image from 3DMM does not achieve photo‐realistic quality on its own, it provides a strong image‐level prior, in parallel with high‐level face semantics, to guide the DDPM for high fidelity image generation. The experimental results demonstrate that our method outperforms state‐of‐the‐art face swapping methods on benchmark datasets in terms of its synthesis quality, and capability to preserve the target face attributes and swap the source face identity.

Solid modelling approach for 3D tolerance analysis of linear dimension applied to planar faces in an assembly

ABSTRACT This paper presents a 3D tolerance analysis approach for linear dimensions applied to planar faces in an assembly. The assembly variations are generated and visualized as an explicit geometrical stack-up of the component variations using the solid modeller Solidworks®. The feature variations are obtained by adapting the geometric solid model of each component, either by offsetting the target planar face or by tilting it within the tolerance zone. A concept of Oriented Minimum Bounding Box (OMBB) is introduced to generate individual component variations with any generalized shape of the target planar face. The analysis of the OMBB extents, the tilting angles and the corresponding pivot points has revealed symmetry in these data. Rigorous mathematical formulations have been implemented in this study to handle the general case of large and small displacements. An approach is suggested to evaluate the functional dimensions, the target face’s centroid and normal for each assembly variation. Functional dimensions of the assembly variations obtained by the software ‘3DCS Variation Analyst’ are found to deviate from those obtained by the proposed approach by up to 40% of the assembly tolerance size. 3DCS tool has also failed to detect out-of-specification assembly variations, which were identified by the proposed approach.

Differential modulation on neural activity related to flankers during face processing: A visual crowding study

In this visual crowding study, we manipulated the perceivability of a central crowded face (a fearful or a neutral face) by varying the similarity between the central face and the surrounding flanker stimuli. We presented participants with pairs of visual clutters and recorded their electroencephalography during an emotion judgement task. In an upright flanker condition where both the central target face and flanker faces were upright faces, participants were less likely to report seeing the target face, and their P300 was weakened, compared to a scrambled flanker condition where scrambled face images were used as flankers. Additionally, at ∼ 120 ms post-stimulus, a posterior negativity was found for the upright compared to scrambled flanker condition, however only for fearful face targets. We concluded that early neural responses seem to be affected by the perceptual characteristics of both target and flanker stimuli whereas later-stage neural activity is associated with post-perceptual evaluation of the stimuli in this visual crowding paradigm.

Evaluating construction parameters of HB and CD methods for super large section tunnel: a case study

The double sidewall guide pit and centre cross diagram (CRD) methods are often used for the construction of large-section tunnels through water-rich fault fracture zones due to their long construction time and high construction cost. To shorten the construction period and save costs, the top heading and benching method (HB) and centre diaphragm (CD) can be chosen for construction. The construction parameters of the step and CD methods are optimized to ensure the surrounding rock stability and tunnel safety. By relying on the Tongzi Tunnel, we simulate the excavation of different step heights in the construction of the top heading and benching method (HB) and CD methods through numerical simulation, the laws of tunnel vault settlement, and changes in the surrounding rock stress, initial support axial force, bending moment and safety factor. The study shows that as the height of the upper step increases, the settlement of the vault and the convergence of the periphery increase, the initial support safety factor decreases, and the plastic zone of the surrounding rock increases at 30 m from the target face. The step height for the top heading and benching method (HB) of construction is optimized as follows. The ratios of the upper, middle and lower step heights are 0.45H, 0.35H and 0.2H, respectively. The CD method construction step height is optimized to the left (right) upper and left (right) lower step height ratios of 0.5H and 0.5H, respectively.

Event-related potential correlates of implicit processing of own- and other-race faces in children

Human adults typically experience difficulties in recognizing and discriminating individual faces belonging to racial groups other than their own. The origin of this “other-race” effect is set in infancy, but the understanding of its developmental course is fragmented. We aimed to access the mechanisms of the other-race effect in childhood by unraveling the neural time course of own- and other-race face processing during a masked priming paradigm. White 6- and 7-year-old children (N = 19) categorized fully visible Asian (other-race) or White (own-race) target faces according to gender. Target faces were preceded by masked same-identity or different-identity prime faces, matching the target for race and gender. We showed an early priming effect on the N100 component, with larger amplitude to different-face pairs than to same-face pairs, and a later race effect on the N200 component, with larger amplitude to own-race face pairs than to other-race face pairs. Critically, race did not interact with priming at any processing stage (P100, N100, P200, N200, or P300). Our results suggest that race could have a temporally limited impact on face processing and that the implicit and unconscious identity processing of own- and other-race faces could be similar in 6- and 7-year-olds, depicting an immature other-race effect during childhood.

Explicit face memory abilities are positively related to the non-intentional encoding of faces: Behavioral and ERP evidence

Individual differences in face memory abilities have been shown to be related to individual differences in brain activity. The present study investigated brain-behavior relationships for the N250 component in event-related brain potentials, which is taken as a neural sign of face familiarity. We used a task in which a designated, typical target face and several (high- and low-distinctive) nontarget faces had to be distinguished during multiple presentations across a session. Separately, face memory/recognition abilities were measured with easy versus difficult tasks. We replicated an increase of the N250 amplitude to the target face across the session and observed a similar increase for the non-target faces, indicating the build-up of memory representations also for these faces. On the interindividual level, larger across-session N250 amplitude increases to low-distinctive non-target faces were related to faster face recognition as measured in an easy task. These findings indicate that non-intentional encoding of non-target faces into memory is associated with the swift recognition of explicitly learned faces; that is, there is shared variance of incidental and intentional face memory.

Study of image sensors for enhanced face recognition at a distance in the Smart City context

Smart monitoring and surveillance systems have become one of the fundamental areas in the context of security applications in Smart Cities. In particular, video surveillance for Human Activity Recognition (HAR) applied to the recognition of potential offenders and to the detection and prevention of violent acts is a challenging task that is still undergoing. This paper presents a method based on deep learning for face recognition at a distance for security applications. Due to the absence of available datasets on face recognition at a distance, a methodology to generate a reliable dataset that relates the distance of the individuals from the camera, the focal length of the image sensors and the size in pixels of the target face is introduced. To generate the extended dataset, the Georgia Tech Face and Quality Dataset for Distance Faces databases were chosen. Our method is then tested and applied to a set of commercial image sensors for surveillance cameras using this dataset. The system achieves an average accuracy above 99% for several sensors and allows to calculate the maximum distance for a sensor to get the required accuracy in the recognition, which could be crucial in security applications in smart cities.

ISFB-GAN: Interpretable semantic face beautification with generative adversarial network

Face beautification is an emerging application widely used in the entertainment industry. Previous studies have proved that Generative Adversarial Networks (GANs) can successfully build the mapping from a general face to a target face. However, their facial style code in the latent space is taken as a black box with a lack of semantics interpretation. They usually fail to consider the geometric semantic information, and cannot focus on the facial beauty-related regions. Moreover, most of them change the background information. To overcome these problems, we propose a novel Interpretable Semantic Generative Adversarial Network (ISFB-GAN) for face beautification. First, we design a new dual-branch facial style encoder to factorize the abundant semantic information of the reference face into the facial geometry style codes and the appearance style codes in the latent space. Then, a novelty facial attention generator is proposed to preserve more background information in the generated images. In the attention generator, the facial style attention is designed to focus on the most discriminative facial region; the background attention is designed to retain the background information. Finally, we propose a hybrid loss function including geometric loss, orthogonality loss, segmentation loss, style reconstruction loss, perceptual loss, etc., to enhance the learning capacity and improve the visual quality. The combination of the geometric loss and the orthogonality loss is deployed to guarantee the disentanglement between the semantic information. Extensive experiments conducted on the CelebAMask-HQ dataset and the SCUT-FBP5500 dataset, demonstrate that ISFB-GAN performs better than the state-of-the-art methods in facial attractiveness enhancement.

Face identity and expression consistency for game character face swapping

Customizing the appearance of game characters according to individual preferences is an important application in the gaming industry. The traditional solutions such as manual editing within the game engine require much time and some professional experience. To address this problem, this paper proposes a novel face-swapping framework to swap real faces onto game characters while maintaining the art style of the game. This approach helps players speed up the character customization process, as they can input any face and quickly see the results of swapping it onto the character. Specifically, our framework extracts a more robust identity embedding from compound face identity models. Then the identity embedding is sent into AdaIN for feature fusion and changes the identity attribute for the output of the decoder. To maintain the expression consistency between the swapped and target faces, we utilize a novel expression embedding loss which effectively constrains the fine-grain expression similarity. To reduce the cross-domain gap between human and game faces, we also construct a game face dataset and propose to use fine-tuning to improve the image quality of cross-domain face swapping. Extensive qualitative and quantitative experiments indicate that our method achieves leading swapping results in both pure natural human faces and game faces datasets.

Immediate and Differential Response to Emotional Stimuli Associated With Transcranial Direct Current Stimulation for Depression: A Visual-Search Task Pilot Study

ObjectivesWhen administered in repeated daily doses, transcranial direct current stimulation (tDCS) directed to the prefrontal cortex has cumulative efficacy for the treatment of depression. Depression can be marked by altered processing of emotionally salient information. An acute marker of response to tDCS may be measured as an immediate change in emotional information processing. Using an easily administered web-based task, we tested immediate changes in emotional information processing in acute response to tDCS in participants with and without depression. Materials and MethodsWe enrolled n = 21 women with mild-to-moderate depression and n = 20 controls without depression to complete a web-based visual search task before and after 30 minutes of tDCS directed to the prefrontal cortex. The timed task required participants to identify a target face among arrays showing sad, neutral, or mixed (distractor) expressions. ResultsAt baseline, as predicted, the participants with depression differed from those without in emotional processing speed (mean z score difference −0.66 ± 0.27, p = 0.022) and accuracy in identifying sad stimuli (error rate: 4.4% vs 1.8%, p = 0.039). In response to tDCS, the participants with depression became significantly faster on the distractor condition (pre- vs post-tDCS z scores: −0.45 ± 0.65 vs −0.85 ± 0.65, p = 0.009), suggesting a specific reduction in bias toward negative emotional information. In response to tDCS, the depressed group also had significant improvements in self-reported mood (increased happy, decreased sad and anxious mood). ConclusionsParticipants with depression vs those without were differentiated by their performance of the visual search task at baseline and in response to tDCS. Given that measurable effects on depression scales may require weeks of tDCS treatments, acute change in emotional information processing can serve as an easily obtainable marker of depression and its response to tDCS. Clinical Trial RegistrationThe Clinicaltrials.gov registration number for the study is NCT05188248.

Why is it difficult for children and adults to follow a person's eye gaze in polynomial social relationships with compound audio-visual stimuli: An eye-tracking study.

Acquiring triadic social interactions could facilitate language and communication skills in early infancy. However, studies have rarely investigated polynomial relationships, defined as relationships among the self, two or more people, and objects. During the development from a child to an adult, the responsiveness to a preferred stimulus modality changes from visual to auditory dominance. Nevertheless, how people observe compound visual stimuli in polynomial social relationships and why it is difficult to ignore auditory cues remain unclear. Moreover, there is a need to identify differences between children's and adults' observing latencies in the time to the first fixation when detecting a stimulus. This study examined whether participants (24 adults and 19 children) demonstrated similar gaze patterns under triadic and polyadic conditions. The participants observed a target visual stimulus looked at by a face stimulus while we presented spoken names, either congruent or incongruent with the target visual stimulus. The results indicated that when the number of people in social relationships increased, children and adults decreased fixations on the target face and the stimulus and showed a shorter mean fixation duration on the face. Moreover, children had longer latencies and more fixation errors for the target stimulus, which might reflect children's difficulties in communicating with others. We expect that understanding children's communication transition from triadic to polynomial social relationships with audio-visual stimulus congruencies would facilitate understanding language development and social communication patterns.

The role of visual imagery in face recognition and the construction of facial composites. Evidence from Aphantasia

People with aphantasia have a markedly impaired ability to form visual images in the mind's eye. Here, by testing people with and without aphantasia, we examine the relationship between visual imagery and face processing. We show that aphantasics have weaker face recognition than people with visual imagery, using both self-report (Prosopagnosia Index) and behavioural measures (Cambridge Face Memory Test). However, aphantasics nonetheless have a fully intact ability to construct facial composites from memory (i.e., composites produced using EFIT6 by aphantasics and imagers were rated as equally accurate in terms of their resemblance to a target face). Additionally, we show that aphantasics were less able than imagers to see the resemblance between composites and a target face, suggestive of potential issues with face matching (perception). Finally, we show that holistic and featural methods of composite construction using EFIT6 produce equally accurate composites. Our results suggest that face recognition, but not face composite construction, is facilitated by the ability to represent visual properties as ‘pictures in the mind’. Our findings have implications for the study of aphantasia, and also for forensic settings, where face composite systems are commonly used to aid criminal investigations.

The “Happy Face Killer” in the eyes of the beholder: Relational encoding of facial emotions in context influences trustworthiness attributions

Research on face perception has established that faces surrounded by threatening contexts are perceived as less trustworthy. Moreover, recent studies revealed that such a face-context integration effect is moderated by the nature of the relational qualifier connecting the face and the context: presenting a face as belonging to either the perpetrator or the victim of the threatening context changes its perceived trustworthiness. Here, we asked whether relational qualifiers can also be extracted from subtle facial cues. In two preregistered studies (N = 225), we tested whether face-context integration is qualified by facial emotions. In Experiment 1, faces appeared with either happy or fearful emotional expressions in threatening contexts (vs. no context). Facial emotions moderated face-context integration: The negative impact of contextual threat on attributed trustworthiness showed stronger for happy than fearful faces. In Experiment 2, participants judged the emotional stimuli on both trustworthiness and smartness. Emotions altered face-context integration when judging trustworthiness but did not alter smartness judgments. Moreover, participants' tendency to judge happy faces in threatening contexts as less trustworthy correlated with their belief that the target face belonged to the criminal on the scene. The importance of considering relational encoding when studying person perception is discussed.

Rethinking one-shot face reenactment: A spatial–temporal reconstruction view

One-shot face reenactment aims to animate a source facial image into a different pose and expression provided by a driving image. Despite recent advancements showing promising results, challenges still exist in accurately transferring facial expressions and being robust to large-scale head movements. Additionally, the current method’s lack of interpretability in the network’s generalization remains a concern. We propose a novel and interpretable framework for one-shot face reenactment based on spatial–temporal reconstruction views to address these challenges. To achieve precise transfer of facial expressions and intuitive control over the generated images, we have opted for the powerful 3D Morphable Face Model (3DMM) as an immediate representation. Transferring motion parameters such as pose and expression from the driving image to the target face allows us to generate a 3D reconstructed face image that allows explicit control. To enhance the robustness of motion reconstruction, we introduce the concept of dense pseudo-temporal flow and propose a cross-modal cascade temporal motion reconstruction network to predict it. Our experimental results demonstrate its effectiveness in improving the robustness of motion reconstruction, especially for large-scale head movements. In addition, to further improve the generated image quality, we propose the motion-modulative residual fusion block as the information path to transfer the motion information from the motion reconstruction network to the local texture reconstruction network. Finally, we discussed the generalization ability of our framework based on its interpretable generation process. Extensive experiments demonstrate that our framework can generate highly realistic facial images and can be comparable with state-of-the-art methods.

The background assimilation effect: Facial emotional perception is affected by surrounding stimuli.

The perception of facial emotion is not only determined by the physical features of the face itself but also be influenced by the emotional information of the background or surrounding information. However, the details of such effect are not fully understood. Here, the authors tested the perceived emotion of a target face surrounded by stimuli with different levels of emotional valence. In Experiment 1, four types of objects were divided into three groups (negative, unpleasant flowers and unpleasant animals; mildly negative (neutral), houses; positive, pleasant flowers). In Experiment 2, three groups of surrounding faces with different social-emotional valence (negative, neutral, and positive) were formed with the memory of affective personal knowledge. The data from two experiments showed that the perception of facial emotion can be influenced and modulated by the emotional valence of the surrounding stimuli, which can be explained by assimilation: the positive stimuli increased the valence of a target face, while the negative stimuli comparatively decreased it. Furthermore, the neutral stimuli also increased the valence of the target, which could be explained by the social positive effect. Therefore, the process of assimilation is likely to be a high-level emotional cognition rather than a low-level visual perception. The results of this study may help us better understand face perception in realistic scenarios.

Conflict and control in cortical responses to inconsistent emotional signals in a face-word Stroop.

Social communication is fraught with ambiguity. Negotiating the social world requires interpreting the affective signals we receive and often selecting between channels of conflicting affective information. The affective face-word Stroop (AFWS) provides an experimental paradigm which may identify cognitive-affective control mechanisms underpinning essential social-affective skills. Initial functional magnetic resonance imaging (fMRI) study of the AFWS identified right amygdala as driving this affective conflict and left rostral anterior cingulate cortex (rACC) as the locus of conflict control. We employed electroencephalogram (EEG) and eLORETA source localization to investigate the timing, location, and sequence of control processes when responding to affective conflict generated during the AFWS. However we designated affective word as the response target and affective face as the distractor to maximize conflict and control effects. Reaction times showed slowed responses in high vs. low control conditions, corresponding to a Rabbitt type control effect rather than the previously observed Grattan effect. Control related activation occurred in right rACC 96-118 ms post-stimulus, corresponding to the resolution of the P1 peak in the Visual Evoked Potential (VEP). Face distractors elicit right hemisphere control, while word distractors elicit left hemisphere control. Low control trials require rapid "booting up" control resources observable through VEPs. Incongruent trial activity in right fusiform face area is suppressed 118-156 ms post stimulus corresponding to onset and development of the N170 VEP component. Results are consistent with a predicted sequence of rapid early amygdala activation by affective conflict, then rACC inhibition of amygdala decreasing facilitation of affective face processing (however, amygdala activity is not observable with EEG).

FlowFace: Semantic Flow-Guided Shape-Aware Face Swapping

In this work, we propose a semantic flow-guided two-stage framework for shape-aware face swapping, namely FlowFace. Unlike most previous methods that focus on transferring the source inner facial features but neglect facial contours, our FlowFace can transfer both of them to a target face, thus leading to more realistic face swapping. Concretely, our FlowFace consists of a face reshaping network and a face swapping network. The face reshaping network addresses the shape outline differences between the source and target faces. It first estimates a semantic flow (i.e. face shape differences) between the source and the target face, and then explicitly warps the target face shape with the estimated semantic flow. After reshaping, the face swapping network generates inner facial features that exhibit the identity of the source face. We employ a pre-trained face masked autoencoder (MAE) to extract facial features from both the source face and the target face. In contrast to previous methods that use identity embedding to preserve identity information, the features extracted by our encoder can better capture facial appearances and identity information. Then, we develop a cross-attention fusion module to adaptively fuse inner facial features from the source face with the target facial attributes, thus leading to better identity preservation. Extensive quantitative and qualitative experiments on in-the-wild faces demonstrate that our FlowFace outperforms the state-of-the-art significantly.