Pattern Recognition最新文献

{"title":"REHair: Efficient hairstyle transfer robust to face misalignment","authors":"Yiwen Xu ,&nbsp;Liping Ling ,&nbsp;Qingxu Lin ,&nbsp;Ying Fang ,&nbsp;Tiesong Zhao","doi":"10.1016/j.patcog.2025.111533","DOIUrl":"10.1016/j.patcog.2025.111533","url":null,"abstract":"<div><div>Hairstyle transfer is challenging due to intricate nature of hairstyles. In particular, face misalignment leads to distortion or deformation of the transferred hairstyle. To address this issue, we propose a Robust and Efficient Hairstyle transfer (REHair) framework, which comprises three stages: adaptive angle alignment, adaptive depth alignment, and efficient hairstyle editing. Firstly, we perform head pose estimation and adjust the facial rotation angle based on the latent code, thus ensuring consistent facial orientation between the face image and the hairstyle reference image and preventing hair shape and texture loss from iterative optimization methods. Secondly, we employ monocular depth estimation to predict the face depth of both images and perform adaptive depth alignment, ensuring the preservation of more hairstyle details. Finally, we propose a fast image embedding algorithm and integrate it with the latent code, significantly reducing the image embedding time in StyleGAN2. This adaptation enables REHair to be suitable for real-time applications. Quantitative and qualitative evaluations on the FFHQ and CelebA-HQ dataset demonstrate that REHair achieves state-of-the-art performance by successfully transferring hairstyles between images with different poses. The proposed method significantly reduces image embedding time while preserving image quality, and effectively addresses challenges associated with sub-optimal photography conditions and slow generation speed. Source code avaliable at <span><span>https://github.com/fdwxfy/REHair</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49713,"journal":{"name":"Pattern Recognition","volume":"164 ","pages":"Article 111533"},"PeriodicalIF":7.5,"publicationDate":"2025-03-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143592924","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Behavior capture guided engagement recognition","authors":"Yijun Bei ,&nbsp;Songyuan Guo ,&nbsp;Kewei Gao ,&nbsp;Zunlei Feng","doi":"10.1016/j.patcog.2025.111534","DOIUrl":"10.1016/j.patcog.2025.111534","url":null,"abstract":"<div><div>Engagement recognition aims to assess an individual’s involvement in various activities, which is essential in fields like education, healthcare, and driving. However, existing methods often suffer from performance degradation due to excessive data and distractions. In this paper, we introduce a novel model, the Behavior Capture-guided Transformer (BCTR). One of its key innovations lies in the proposed architecture for extracting regional features. Specifically, BCTR employs three independent class tokens to capture regional features – ocular, head, and trunk – from image sequences. These features are then used to model the dynamic streams of these regions for video-based engagement recognition. Another unique innovation of BCTR is its ability to mimic the observational techniques used by human teachers. By leveraging both frame-level and video-level class tokens, the model uses dual branches to detect both static and dynamic disengagement behaviors. This approach not only enables BCTR to achieve superior performance – 64.51% accuracy on the DAiSEE dataset and 0.0602 MSE loss on the EmotiW-EP dataset – but also enhances the interpretability of engagement levels by identifying these disengagements.</div></div>","PeriodicalId":49713,"journal":{"name":"Pattern Recognition","volume":"164 ","pages":"Article 111534"},"PeriodicalIF":7.5,"publicationDate":"2025-03-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143592926","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Meta-learning of pseudo force field generation and estimation for enhancing 3D molecular property prediction","authors":"Yufei Luo,&nbsp;Heran Yang,&nbsp;Jian Sun","doi":"10.1016/j.patcog.2025.111531","DOIUrl":"10.1016/j.patcog.2025.111531","url":null,"abstract":"<div><div>Learning Energy-based Model via 3D molecular denoising has been shown to be effective in pretraining the 3D molecular representation. However, existing works carry out denoising in task-agnostic manner, causing inevitable domain gap with the downstream tasks. To overcome this issue, we introduce a task-aware pretraining framework, dubbed Mol-MFFGE, for adapting the energy-based pretraining to downstream tasks in meta-learning approach. In this framework, we design learnable pretraining tasks as generating and estimating pseudo force fields. This is achieved by proposing a learnable noise transformation module to generate the noisy motions of atoms and the model is pretrained to estimate them. These tasks are taken as the auxiliary self-supervised training tasks and learned with the downstream task jointly, formulated as a bi-level meta-learning optimization problem. Based on such an approach, the force field generation and estimation tasks are meta-learned to enhance the downstream tasks for molecular property prediction. Extensive experiments are conducted on three molecular property prediction datasets, and results demonstrate performance improvement over the state-of-the-art 3D molecular pretrained models.</div></div>","PeriodicalId":49713,"journal":{"name":"Pattern Recognition","volume":"164 ","pages":"Article 111531"},"PeriodicalIF":7.5,"publicationDate":"2025-03-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143592929","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"ADGaze: Anisotropic Gaussian Label Distribution Learning for fine-grained gaze estimation","authors":"Duantengchuan Li ,&nbsp;Shutong Wang ,&nbsp;Wanli Zhao ,&nbsp;Lingyun Kang ,&nbsp;Liangshan Dong ,&nbsp;Jiazhang Wang ,&nbsp;Xiaoguang Wang","doi":"10.1016/j.patcog.2025.111536","DOIUrl":"10.1016/j.patcog.2025.111536","url":null,"abstract":"<div><div>Gaze estimation technology is crucial for enhancing the effectiveness and safety of applications in human–computer interaction, intelligent driving, virtual reality, and medical diagnosis. With advancements in deep learning, gaze estimation methods using deep neural networks have been extensively researched and applied. However, existing methods have yet to address the anisotropic characteristics of eye features. Based on the discovered anisotropic characteristics, we propose an Anisotropic Gaussian Label Distribution Learning Network for Gaze Estimation (ADGaze). ADGaze is capable of catching neighboring information by taking advantage of coarse-to-fine methodology and the anisotropic soft label construct. The coarse-to-fine framework initially performs classification tasks for gaze estimation, grouping gaze images with small variations into the same category, followed by regression tasks for each category. The construction of anisotropic Gaussian label distributions adopts methods based on data statistics and feature similarity. Extensive experimentation on public datasets has been carried out to substantiate the efficacy of this model. Our code is publicly available at <span><span>https://github.com/dacilab/ADGaze</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49713,"journal":{"name":"Pattern Recognition","volume":"164 ","pages":"Article 111536"},"PeriodicalIF":7.5,"publicationDate":"2025-03-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143610799","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Bridging domain spaces for unsupervised domain adaptation","authors":"Jaemin Na ,&nbsp;Heechul Jung ,&nbsp;Hyung Jin Chang ,&nbsp;Wonjun Hwang","doi":"10.1016/j.patcog.2025.111537","DOIUrl":"10.1016/j.patcog.2025.111537","url":null,"abstract":"<div><div>Unsupervised Domain Adaptation (UDA) aims to transfer knowledge obtained from a labeled source domain to an unlabeled target domain, facing challenges due to domain shift—significant discrepancies in data distribution that impair model performance when applied to unseen domains. While recent approaches have achieved remarkable progress in mitigating these domain shifts, the focus remains on direct adaptation strategies from source to target domains. However, when the gap between the source and target domains is too substantial, directly aligning their distributions becomes increasingly difficult. Pseudo-labeling, a common strategy in direct adaptation, can further exacerbate this issue when the domain shift is severe. In such cases, incorrect pseudo-labels are likely to propagate through the adaptation process, leading to degraded performance and unstable training. Effective adaptation thus requires methods that can address these challenges by improving the reliability of pseudo-labels or reducing dependency on them. To address this challenge, we propose a novel approach that effectively alleviates domain shift by leveraging intermediate domains as bridges between the source and target domains. Specifically, we introduce a fixed ratio-based mixup to generate distinct intermediate domains between the source and target domains. By training on these augmented domains, we construct source-dominant and target-dominant models that possess distinct strengths and weaknesses, enabling us to implement effective complementary learning strategies. Furthermore, we enhance our fixed ratio-based mixup with uncertainty-aware learning, which addresses not only the image-level space but also the feature space, aiming to reduce the uncertainty at the most critical points within these spaces. Finally, we integrate confidence-based learning strategies, including bidirectional matching with high-confidence predictions and self-penalization with low-confidence predictions. Our extensive experiments on seven public benchmarks, including both single-source and multi-source scenarios, demonstrate the effectiveness of our method in UDA tasks.</div></div>","PeriodicalId":49713,"journal":{"name":"Pattern Recognition","volume":"164 ","pages":"Article 111537"},"PeriodicalIF":7.5,"publicationDate":"2025-03-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143610971","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"AFIFC: Adaptive fuzzy neighborhood mutual information-based feature selection via label correlation","authors":"Lin Sun ,&nbsp;Feng Xu ,&nbsp;Weiping Ding ,&nbsp;Jiucheng Xu","doi":"10.1016/j.patcog.2025.111577","DOIUrl":"10.1016/j.patcog.2025.111577","url":null,"abstract":"<div><div>Existing feature selection schemes do not comprehensively consider correlation between features and labels and between labels, and certain neighborhood radius affects the prediction accuracy of multilabel classification. To solve these deficiencies, this paper develops an adaptive fuzzy neighborhood mutual information-based feature selection scheme via label correlation. Firstly, to study different distribution structures of multilabel data, the standard Euclidean distance as classification interval is employed to construct adaptive fuzzy neighborhood radius. Adaptive fuzzy neighborhood similarity relation and fuzzy neighborhood granule will be presented via difference between samples for features. Uncertainty measures via fuzzy neighborhood entropy can be developed. Secondly, to select features strongly associated with labels, adaptive fuzzy neighborhood mutual information measures this correlation between candidate features and labels, and the correlation between features and labels relative to those selected features is computed by mutual information. Then discriminant function of correlation is provided. Thirdly, to improve efficacy of multilabel classification, adaptive fuzzy neighborhood granules are employed to study the membership degree of labels. To assess the correlation between labels, Jaccard similarity and adaptive fuzzy neighborhood mutual information are combined, and to reflect this internal correlation between label and label set, the correlation ratio is studied. Finally, maximum relevance between the candidate features and labels and minimum redundancy between features are calculated, and then a new multilabel feature selection scheme is provided to acquire this best feature subset. Experiments on 12 datasets show the efficacy of this designed scheme in several evaluation metrics.</div></div>","PeriodicalId":49713,"journal":{"name":"Pattern Recognition","volume":"164 ","pages":"Article 111577"},"PeriodicalIF":7.5,"publicationDate":"2025-03-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143629137","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Multi-Agent policy gradients with dynamic weighted value decomposition","authors":"Shifei Ding ,&nbsp;Xiaomin Dong ,&nbsp;Jian Zhang ,&nbsp;Lili Guo ,&nbsp;Wei Du ,&nbsp;Chenglong Zhang","doi":"10.1016/j.patcog.2025.111576","DOIUrl":"10.1016/j.patcog.2025.111576","url":null,"abstract":"<div><div>In real-world multi-agent systems, multiple agents need to coordinate with other agents due to some limitations of observation and communication ability. Multi-agent policy gradient methods recently have witnessed vigorous progress in such challenging settings. However, multi-agent policy gradient methods have scalability and credit assignment issues due to the centralized critic. To solve these issues, a novel <strong>D</strong>ynamic Weighted QMI<strong>X</strong> Based <strong>M</strong>ulti-Agent Policy Gradients (DXM) is proposed in this paper, where the idea of dynamic weighted value decomposition is introduced into the framework of multi-agent actor-critic. Based on this idea, the proposed DXM approach has a more general decomposition on centralized critic than existing value decomposition methods, which address the scalability and credit assignment issue in both continuous and discrete action spaces. Briefly, in the presented DXM, deep deterministic policy gradient is employed to learn policies and a single centralized but factored critic, which can decompose the dynamic weighted nonlinear nonmonotonic summation of individual value functions. Empirical evaluations on the discrete action space environment StarCraft multi-agent challenge benchmark and the continuous action space environment continuous predator-prey benchmark show that the DXM approach successfully addresses the scalability and credit allocation issues. DXM significantly outperforms other baselines, with an average win rate improvement of &gt;15 %.</div></div>","PeriodicalId":49713,"journal":{"name":"Pattern Recognition","volume":"164 ","pages":"Article 111576"},"PeriodicalIF":7.5,"publicationDate":"2025-03-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143610800","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"CLEAR: Cross-Transformers With Pre-Trained Language Model for Person Attribute Recognition and Retrieval","authors":"Doanh C. Bui ,&nbsp;Thinh V. Le ,&nbsp;Ba Hung Ngo ,&nbsp;Tae Jong Choi","doi":"10.1016/j.patcog.2025.111486","DOIUrl":"10.1016/j.patcog.2025.111486","url":null,"abstract":"<div><div>Person attribute recognition and attribute-based person retrieval are two core human-centric tasks. In the recognition task, the challenge lies in identifying attributes based on a person’s appearance, while the retrieval task involves searching for matching persons using attribute-based queries. In this paper, we present <span>CLEAR</span>, a unified network designed to address both tasks. We leverage our C<span><math><msup><mrow></mrow><mrow><mn>2</mn></mrow></msup></math></span>T-Net, a strong Cross-Transformers backbone that achieved state-of-the-art performance in the person attribute recognition task during the UPAR Challenge 2024, to extract visual embeddings. We then adapt it for the attribute-based person retrieval task.To extend its capabilities for the attribute-based person retrieval task, we construct pseudo-textual descriptions for attribute queries, leverage a pretrained language model to generate language-rich feature embeddings, and introduce an effective training strategy, which involves finetuning only a few additional parameters in the form of adapters to produce visual and query embeddings within the retrieval space. As the visual embeddings extracted by C<span><math><msup><mrow></mrow><mrow><mn>2</mn></mrow></msup></math></span>T-Net are highly discriminative, they align well with the proposed query embeddings during the finetuning process, facilitating improved retrieval performance.The unified <span>CLEAR</span>, model is evaluated on five benchmarks: PETA, PA100K, Market-1501, RAPv2, and UPAR2024, achieving state-of-the-art or competitive results for both tasks. Notably, it ranks as the top performer on the large-scale UPAR2024 dataset, specifically designed to test domain generalizability in real-world scenarios where test samples differ from training samples.</div></div>","PeriodicalId":49713,"journal":{"name":"Pattern Recognition","volume":"164 ","pages":"Article 111486"},"PeriodicalIF":7.5,"publicationDate":"2025-03-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143610802","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"TLR-3DRN: Unsupervised single-view reconstruction via tri-layer renderer","authors":"HaoYu Guo ,&nbsp;Ying Li ,&nbsp;Chunyan Deng","doi":"10.1016/j.patcog.2025.111568","DOIUrl":"10.1016/j.patcog.2025.111568","url":null,"abstract":"<div><div>Single-view three-dimensional (3D) reconstruction is a challenging task in computer vision, focusing on reconstructing 3D objects from a single image. Existing single-view object reconstruction approaches typically rely on viewpoints, silhouettes, multiple views of the same instance, and strategy-specific priors, which are difficult to obtain in the wild. To address this issue, we propose a novel end-to-end single-view reconstruction method based on a tri-layer renderer, named the Tri-Layer Renderer-based 3D Reconstruction Network (TLR-3DRN). TLR-3DRN recovers 3D structures from original image collections without requiring additional supervision, assumptions, or priors. In particular, TLR-3DRN employs a tri-layer renderer that enables the model to extract more 3D details from unprocessed image data. To obtain an optimizable interlayer, we developed a robust interlayer generation network based on a nonparametric memory bank. Notably, we designed a joint optimization strategy for the overall framework. Additionally, a shape and texture consistency loss based on image–text models is proposed to enhance the optimization process. Owing to the aforementioned proposed modules, TLR-3DRN can achieve high-quality, diverse-category reconstruction under completely unsupervised conditions. TLR-3DRN is validated on synthetic datasets and real-world datasets. Experimental results demonstrate that TLR-3DRN outperforms state-of-the-art unsupervised and two-dimensional supervised methods, achieving performance comparable to 3D supervised methods.</div></div>","PeriodicalId":49713,"journal":{"name":"Pattern Recognition","volume":"164 ","pages":"Article 111568"},"PeriodicalIF":7.5,"publicationDate":"2025-03-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143576953","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"An efficient approach for finger vein verification to solving the biometric recognition technique","authors":"Dharmalingam Muthusamy ,&nbsp;Rakkimuthu Ponnusamy","doi":"10.1016/j.patcog.2025.111563","DOIUrl":"10.1016/j.patcog.2025.111563","url":null,"abstract":"<div><div>Vein authentication is a novel biometric method to authenticate a person's individuality. The conventional biometric technique employs shape images and exact segments of finger veins for the verification process. The proposed deep belief structure model aims to improve verification accuracy using a novel Anisotropic Filtered Stromberg Feature Transform based on Tucker's Congruence Deep Belief Structure Learning (AFSFT-TCDBSL) technique. The main aim of the AFSFT-TCDBSL technique is to improve verification accuracy and minimize time consumption. The proposed AFSFT-TCDBSL technique comprises one input layer, three hidden layers, and one output layer. The numbers of images are collected in the input layer, and the input images are pre-processed using anisotropic diffusion filtering in the first hidden layer. Then the pre-processed input images are sent to the next layer, where the feature extraction process is carried out using the Stromberg wavelet transform. Finally, the verification process is performed using Tucker's congruence correlation coefficient. Based on the correlation, the verification results are obtained at the output layer. In this way, accurate finger vein verification is performed with superior accuracy and with a minimum false rate. We performed experimental assessments with different factors, such as the Peak Signal-to-Noise Ratio (PSNR), Finger Vein Verification Accuracy (FVVA), False Positive Rate (FPR), Processing Time (PT), and Feature Extraction Time (FET). The results of the proposed ADFSFT-TCDBSL technique were conducted on 9% of improved peak signal-to-noise ratio and accuracy with a minimum 59% false positive rate and 16% time as well as 19% feature extraction time than the state-of-the-art FVV methods; therefore, it better facilitates the application of real-time finger vein verification.</div></div>","PeriodicalId":49713,"journal":{"name":"Pattern Recognition","volume":"164 ","pages":"Article 111563"},"PeriodicalIF":7.5,"publicationDate":"2025-03-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143610803","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}