Search Results (46)

Facial surface temperature provides useful non-contact information for monitoring dairy-cow health, welfare, heat stress, and reproductive status. Infrared thermography can capture thermal information from facial regions such as the eyes, muzzle, nostrils, and ears; however, infrared images often contain weak texture and indistinct anatomical boundaries, which can hinder reliable cow-head localization during mobile robotic inspection. Visible-light images provide richer structural information but do not contain temperature data. This study developed YOLO11-AFE, a lightweight visible–thermal cow-head detection and heterogeneous-camera association method for quadruped inspection robots. The detector incorporates ADown for lightweight downsampling, C3k2_FE for adaptive feature enhancement, and SPPF_ECA for channel-aware multi-scale representation. An improved Hungarian matching algorithm was subsequently used to establish one-to-one correspondences between cow-head detections in synchronized visible-light and infrared pseudo-colour images. Across three independent runs, YOLO11-AFE achieved precision, mAP@0.5, and mAP@0.5:0.95 values of 97.59 ± 0.20%, 96.37 ± 0.24%, and 70.76 ± 0.51%, respectively, on the visible-light test subset, and 96.11 ± 0.24%, 99.07 ± 0.10%, and 91.50 ± 0.40%, respectively, on the infrared subset. The model required 2.14 million parameters and 5.27 GFLOPs, representing reductions of 17.37% and 18.17%, respectively, relative to YOLO11n. The association method achieved an overall accuracy of 98.11% across 371 ground-truth cow-head pairs in the combined validation and test evaluation. TensorRT FP16 deployment on the Jetson Orin NX achieved 36.71 FPS for the complete core processing pipeline. These results demonstrate that YOLO11-AFE provides an accurate and computationally efficient perception front end for future non-contact facial-temperature monitoring using mobile inspection robots. Full article

Facial expression recognition in the wild is challenged by both noisy supervision and degraded visual evidence: subtle expression cues must be interpreted under blur, contrast changes, image noise, and annotator disagreement. Existing noisy-label FER methods mainly regulate samples, labels, or attention, while frequency information is rarely used to adapt the semantic representation itself. We propose Frequency-Guided Expert Modulation (FARM-FER), which treats local and global frequency descriptors as a control signal rather than an additional classifier input. A joint Haar-DWT and radial-FFT context guides soft routing among nonlinear experts and channel-wise affine recalibration of the semantic feature, while a learned gate combines the two corrections before a lightweight classifier predicts the expression from the refined representation. Across RAF-DB, FER+, and AffectNet under symmetric label noise, with additional evaluations under class-dependent label noise on RAF-DB and native crowd-label ambiguity on FER+, FARM-FER consistently improves matched baselines. At 30% symmetric noise, FARM-FER reaches 89.18% accuracy on RAF-DB, with a 1.6% performance gain over the matched Swin-Tiny baseline; the gains also hold in a controlled ResNet18 reimplementation and in class-sensitive AffectNet evaluation. Measured cost analyses show only modest parameter and FLOP overhead, supporting a lightweight yet effective design in terms of model size and arithmetic cost for noisy-label FER. Full article

We introduce the Probabilistic Multimodal Consistency Index (PMCI), a prototype measure of probabilistic semantic agreement used to quantify how much two modalities share in terms of mutually compatible affective evidence. PMCI does not model feature fusion or emotion classification directly; instead, each modality is mapped to a different probability distribution over learnable latent affective-agreement anchors, and the agreement of those distributions is quantified via the Jensen–Shannon divergence. Consequently, we propose PMCI not as a substitute for a discriminative model of pair matching, but rather as an auxiliary diagnostic index of cross-modal affective agreement. Experiments were conducted on pose-based facial and body keypoint sequences obtained from the RAVDESS and MELD datasets. In the updated RAVDESS setup, we extended the cache of preprocessed keypoint files to include all 24 actors, and we sampled four distinct pose-derived 12-frame crops per source video. This resulted in a total of 11,520 pose-derived windows, with the actor label determining the train/validation/test splits for the RAVDESS dataset. MELD contained 402 filtered pose-derived windows and served as an auxiliary in-the-wild validation setting with additional noise. When applying the updated RAVDESS standard pair-matching, DirectCosine_K0 achieved ROC–AUC = 0.943 and PR–AUC = 0.917, demonstrating that it is indeed the best exact pair-matching baseline and that PMCI-based configurations are not as accurate as DirectCosine_K0 when performing common exact face–body pair discrimination. Under this protocol, PMCI_K8, PMCI_K16, and PMCI_K32 produced ROC–AUC scores of 0.851, 0.805, and 0.887, respectively. Running a one-window-per-source-video experiment with 100 repetitions yielded similar results in the following order: DirectCosine_K0 = 0.942 plus-minus 0.009, PMCI_K16 = 0.807 plus-minus 0.016, and PMCI_K32 = 0.893 plus-minus 0.012, which shows that the results obtained were not the result of only repeated temporal crops. In the DirectCosine-mined hard negative setting, PMCI produced modest diagnostic separation above chance, where Direct+PMCI_K32 yielded ROC–AUC = 0.582. In a suite of shuffle control, permutation control, model initialization control, temperature control, anchor usage control, control latency, and pose perturbation control experiments, the diagnostic stability of PMCI was tested, confirming that PMCI is sensitive to both pose-estimation accuracy and domain shift. Finally, PMCI must be interpreted as a probabilistic diagnostic index of cross-modal affective agreement that is based on pose, not as a general-purpose emotion-recognition system. Full article

Conventional cosmetic human studies rely on pre–post mean comparisons, which have limitations in explaining where and how facial skin changes occur. This pilot single-arm study proposed a privacy-preserving three-dimensional (3D) facial mesh mapping framework and demonstrated its application using an illustrative dataset obtained from participants who used a polydeoxyribonucleotide (PDRN)-containing cosmetic. Twenty-two participants underwent facial skin assessments before and after product use. Conventional analysis included pre–post comparisons of elasticity-related parameters. Additionally, 3D facial images obtained via stereophotogrammetry were converted into de-identified mesh surfaces, spatially aligned between time points, and visualized using color-coded heatmaps. For each participant, the left facial panel displayed changes in a skin hydration permittivity index, while the right panel displayed changes in the R2 gross elasticity parameter (Ua/Uf). Overall mean values tended to increase after product use; however, the 3D visualization revealed heterogeneous spatial patterns undetectable via mean values. This method improved spatial matching, enabled intuitive regional comparison, and reduced privacy concerns by removing identifiable facial features. The privacy-preserving 3D facial mesh mapping (P3DMM) framework may serve as a complementary tool for cosmetic human studies, enabling the generation of structured, de-identified spatial datasets for future skin response research. Full article

Background: Facial motion has been shown to enhance face recognition in typical recognisers and increase attention to diagnostic internal facial features (eyes, nose and mouth), which may support improved recognition. The present study aimed to replicate these effects and assess whether they extend to individuals with developmental prosopagnosia (DP), who show impairments in face recognition and reduced attention to the internal features. Methods: Participants completed a famous face recognition task (Experiment 1) and an unfamiliar old/new recognition task (Experiment 2), with both static and moving stimuli. Eye movements were recorded to assess visual attention. In each experiment, two analyses were conducted: a replication analysis in a neurotypical sample (Experiment 1: n = 49; Experiment 2: n = 51), and a separate comparison of 14 individuals with DP with a subset of 16 age-matched controls. Results: Across both experiments, recognition accuracy was higher for moving than static faces in both the control and DP samples, and participants directed a greater proportion of visual attention to internal facial features when faces were presented in motion. However, compared to age-matched controls, individuals with DP showed differences in the allocation of attention to the internal facial features. Conclusions: The findings replicate evidence that facial motion enhances recognition and increases attention to the internal facial features in typical recognisers and extend these effects to individuals with DP. Although individuals with DP benefit from motion, differences in the distribution of attention across internal features remain, suggesting that motion alters but does not normalise face-scanning strategies. Full article

A novel facial landmark localization method, termed TSDM, is proposed by integrating T-shaped features with the Supervised Descent Method (SDM). Facial landmark localization is critical for driver fatigue and attention detection in intelligent cockpits. Traditional methods lack accuracy and robustness in complex in-cabin environments such as varying illumination and head pose changes, while deep learning approaches are computationally expensive on resource-constrained vehicle platforms. The T-shaped feature well matches facial geometry and enhances feature representation. T-shaped features are selected via AdaBoost for robust face detection, and SDM is then used to locate 68 facial landmarks. Experiments show that TSDM achieves higher accuracy, lower false-positive rates, and better efficiency than traditional methods, including Haar and LBPH. It also exhibits stronger robustness and better real-time performance than several lightweight deep learning models (such as 3D-aware methods and SAN) on CPU-only platforms, while achieving comparable or higher localization accuracy. Experimental results show that TSDM achieves a face detection rate of 97.43% and a normalized mean error (NME) of 3.4% on standard datasets. The proposed method provides a practical solution for driver state monitoring in resource-limited vehicular environments. Full article

Cross-age face analysis has emerged as an important topic in biometric recognition due to substantial facial appearance variations caused by aging. Nevertheless, most existing approaches primarily focus on face verification (1:1 matching) and frequently rely on explicit age annotations, which limit their applicability in large-scale retrieval scenarios. In this study, large-scale cross-age face retrieval (1:N matching) is investigated, and a Hybrid Metric Learning Framework (HMLF) is proposed to learn age-invariant and retrieval-oriented facial representations without requiring age labels. The proposed framework integrates Additive Angular Margin Loss (ArcFace) with supervised contrastive learning to enhance feature discriminability. Furthermore, a mixed triplet mining strategy is introduced to improve the effectiveness of hard sample selection. A memory bank-based InfoNCE formulation is incorporated to provide a large number of negative samples, and an uncertainty-based adaptive weighting scheme is designed to automatically balance multiple loss components during optimization. To better simulate realistic retrieval scenarios, an extended cross-age retrieval evaluation protocol is established. Extensive experimental results demonstrate that the proposed framework achieves superior retrieval performance across different backbone architectures. The results further provide systematic insights into the influence of backbone design, loss formulation, and optimization strategies on cross-age retrieval accuracy. Full article

Agricultural mechanization and advanced technology have developed significantly in the coffee industry. However, there are still requirements for human laborers to operate, monitor crop health care, and manage production. The integration of advanced technology can significantly enhance the production efficiency and management practices of agricultural enterprises. This paper aims to address these gaps by proposing and implementing a computer vision-based attendance tracking system on mobile platforms that are suitable for the requirements and limitations of agricultural enterprises. First, the face detection process involves interpreting and locating facial structure. Next, the model transforms a photographic image of a human face into digital data based on the unique features and facial structure. We utilize the InsightFace model with the buffalo_l variant, as well as ArcFace with a ResNet backbone, as a facial recognition algorithm. After capturing a facial image, the system conducts a matching process against the existing database to verify identity. Finally, we implement a mobile application prototype on both iOS and Android platforms, ensuring accessibility for farm workers. As a result, our system achieved 95.2% accuracy on the query set, with an average processing time of <200 ms per image (including face detection, embedding extraction, and database matching). The system performs real-time attendance monitoring, automatically recording the entry and exit times of farm workers using facial recognition technology, and enables quick registration of new workers. Our work is expected to enhance transparency and fairness in the human management process, focusing on the coffee farm use case. Full article

In the context of privacy-preserving face recognition systems, entropy plays a crucial role in determining the efficiency and security of computational processes. However, existing schemes often encounter challenges such as inefficiency and high entropy in their computational models. To address these issues, we propose a privacy-preserving face recognition method based on the Face Feature Coding Method (FFCM) and symmetric homomorphic encryption, which reduces computational entropy while enhancing system efficiency and ensuring facial privacy protection. Specifically, to accelerate the matching speed during the authentication phase, we construct an N-ary feature tree using a neural network-based FFCM, significantly improving ciphertext search efficiency. Additionally, during authentication, the server computes the cosine similarity of the matched facial features in ciphertext form using lightweight symmetric homomorphic encryption, minimizing entropy in the computation process and reducing overall system complexity. Security analysis indicates that critical template information remains secure and resilient against both passive and active attacks. Experimental results demonstrate that the facial authentication efficiency with FFCM classification is 4% to 6% higher than recent state-of-the-art solutions. This method provides an efficient, secure, and entropy-aware approach for privacy-preserving face recognition, offering substantial improvements in large-scale applications. Full article

Pilot visual behavior safety assessment is a cross-disciplinary technology that analyzes pilots’ gaze behavior and neurocognitive responses. This paper proposes a multimodal analysis method for pilot visual behavior safety, specifically for cognitive state decoding. This method aims to achieve a quantitative and efficient assessment of pilots’ observational behavior. Addressing the subjective limitations of traditional methods, this paper proposes an observational behavior detection model that integrates facial images to achieve dynamic and quantitative analysis of observational behavior. It addresses the “Midas contact” problem of observational behavior by constructing a cognitive analysis method using multimodal signals. We propose a bidirectional long short-term memory (LSTM) network that matches physiological signal rhythmic features to address the problem of isolated features in multidimensional signals. This method captures the dynamic correlations between multiple physiological behaviors, such as prefrontal theta and chest-abdominal coordination, to decode the cognitive state of pilots’ observational behavior. Finally, the paper uses a decision-level fusion method based on an improved Dempster–Shafer (DS) evidence theory to provide a quantifiable detection strategy for aviation safety standards. This dual-dimensional quantitative assessment system of “visual behavior–neurophysiological cognition” reveals the dynamic correlations between visual behavior and cognitive state among pilots of varying experience. This method can provide a new paradigm for pilot neuroergonomics training and early warning of vestibular-visual integration disorders. Full article

Despite significant advances in gene discovery, the molecular basis of many rare genetic disorders remains poorly understood. The concept of disease modules, clusters of functionally related genes whose disruption leads to overlapping phenotypes, offers a valuable framework for interpreting these conditions. However, identifying such relationships remains particularly challenging in ultra-rare syndromes due to the limited number of documented cases. We hypothesized that AI-based facial phenotyping could aid in identifying shared molecular mechanisms by detecting phenotypic convergence among clinically related syndromes. To test this, we used Schuurs–Hoeijmakers syndrome (SHMS; OMIM #615009), caused by a recurrent de novo variant in PACS1, as a model to explore potential phenotypic and functional associations with PACS2-related disorder (DEE66; OMIM #618067) and WDR37-related disorder (NOCGUS; OMIM #618652). Facial photographs of individuals with SHMS were analyzed using the DeepGestalt and GestaltMatcher algorithms. In addition to consistently recognizing SHMS as a distinct clinical entity, the algorithms frequently matched DEE66 and NOCGUS, suggesting a shared facial gestalt. Binary comparisons further confirmed overlapping craniofacial features among the three disorders. These findings were supported by literature review, indicating clinical overlapping and potential functional associations. Overall, our results confirm the presence of consistent facial similarities among PACS1-, PACS2-, and WDR37-related syndromes and highlight the utility of AI-driven facial phenotyping as a complementary tool for uncovering clinically relevant relationships in ultra-rare genetic disorders. Full article

This research highlights the importance of Emotion Aware Technologies (EAT) and their implementation in serious games to assist children with Autism Spectrum Disorder (ASD) in developing social-emotional skills. As AI is gaining popularity, such tools can be used in mobile applications as invaluable teaching tools. In this paper, a new AI framework application is discussed that will help children with ASD develop efficient social-emotional skills. It uses the Jetpack Compose framework and Google Cloud Vision API as emotion-aware technology. The framework is developed with two main features designed to help children reflect on their emotions, internalise them, and train them how to express these emotions. Each activity is based on similar features from literature with enhanced functionalities. A diary feature allows children to take pictures of themselves, and the application categorises their facial expressions, saving the picture in the appropriate space. The three-level minigame consists of a series of prompts depicting a specific emotion that children have to match. The results of the framework offer a good starting point for similar applications to be developed further, especially by training custom models to be used with ML Kit. Full article

Background/Objectives: Successful discourse relies not only on linguistic but also on prosodic information. Difficulty recognizing emotion conveyed through prosody (receptive affective aprosodia) following right hemisphere stroke (RHS) significantly disrupts communication participation and personal relationships. Growing evidence suggests that damage to white matter in addition to gray matter structures impairs affective prosody recognition. The current study investigates lesion–symptom associations in receptive affective aprosodia during RHS recovery by assessing whether disruptions in distinct white matter structures impact different underlying affective prosody recognition skills. Methods: Twenty-eight adults with RHS underwent neuroimaging and behavioral testing at acute, subacute, and chronic timepoints. Fifty-seven healthy matched controls completed the same behavioral testing, which comprised tasks targeting affective prosody recognition and underlying perceptual, cognitive, and linguistic skills. Linear mixed-effects models and multivariable linear regression were used to assess behavioral performance recovery and lesion–symptom associations. Results: Controls outperformed RHS participants on behavioral tasks earlier in recovery, and RHS participants’ affective prosody recognition significantly improved from acute to chronic testing. Affective prosody and emotional facial expression recognition were affected by external capsule and inferior fronto-occipital fasciculus lesions while sagittal stratum lesions impacted prosodic feature recognition. Accessing semantic representations of emotions implicated the superior longitudinal fasciculus. Conclusions: These findings replicate previously observed associations between right white matter tracts and affective prosody recognition and further identify lesion–symptom associations of underlying prosodic recognition skills throughout recovery. Investigation into prosody’s behavioral components and how they are affected by injury can help further intervention development and planning. Full article

Facial soft tissue reconstruction is an important tool in forensic investigations, especially when conventional identification methods are unsuccessful. This paper presents a digital workflow for facial reconstruction and identity verification using computer vision techniques applied to two forensic cases. The first case involves a cold case from 1993, in which a manual reconstruction by Prof. Helmer was conducted in 1994. We digitally reconstructed the same individual using CAD software (Blender), enabling a direct comparison between manual and digital techniques. To date, the deceased remains unidentified. The second case, from 2021, involved a digitally reconstructed face that was later matched to a missing person through DNA analysis. Here, comparison material was available, including an official photograph. A police officer involved in the case noted a “striking resemblance” between the reconstruction and the photograph. To evaluate this subjective impression, we performed quantitative analyses using three face recognition models (Dlib-based method, VGG-Face, and GhostFaceNet). The models did not indicate significant similarity, highlighting a gap between human perception and algorithmic assessment. These findings suggest that current face recognition algorithms may not yet be fully suited to evaluating reconstructions, which tend to deviate in subtle but critical facial features. To achieve better facial recognition results, further research is required to generate more anatomically accurate and detailed reconstructions that align more closely with the sensitivity of AI-based identification systems. Full article

Face verification and identification traditionally follow a symmetric matching approach, where the same model (e.g., ResNet-50 vs. ResNet-50) generates embeddings for both gallery and query images, ensuring compatibility. However, real-world scenarios often demand asymmetric matching, especially when query devices have limited computational resources or employ heterogeneous models (e.g., ResNet-50 vs. SwinTransformer). This asymmetry can degrade face recognition performance due to incompatibility between embeddings from different models. To tackle this asymmetric face recognition problem, we introduce the Learnable Anchor Embedding (LAE) model, which features two key innovations: the Shared Learnable Anchor and a Light Cross-Attention Mechanism. The Shared Learnable Anchor is a dynamic attractor, aligning heterogeneous gallery and query embeddings within a unified embedding space. The Light Cross-Attention Mechanism complements this alignment process by reweighting embeddings relative to the anchor, efficiently refining their alignment within the unified space. Extensive evaluations of several facial benchmark datasets demonstrate LAE’s superior performance, particularly in asymmetric settings. Its robustness and scalability make it an effective solution for real-world applications such as edge-device authentication, cross-platform verification, and environments with resource constraints. Full article