Signal Processing-Image Communication最新文献

{"title":"Hybrid Feature Extraction and Knowledge Distillation Based Deep Learning Model for Human Activity Recognition System","authors":"Hetal Shah ,&nbsp;Mehfuza S. Holia","doi":"10.1016/j.image.2025.117308","DOIUrl":"10.1016/j.image.2025.117308","url":null,"abstract":"<div><div>This article introduces the Generative Adversarial Network (GAN) framework model, which uses offline knowledge distillation (KD) to move spatio-deep data from a large teacher to a smaller student model. To achieve this, the teacher model named EfficientNetB7 embedded with spatial attention (E2SA) and a multi-layer Gated Recurrent Unit (GRU) is used. A hybrid feature extraction method known as Completed Hybrid Local Binary Pattern (ChLBP) is employed prior to the prediction process. After feature extraction, the hybrid features are parallelly given as input to both teacher and student models. In the teacher model, E2SA extracts both deep and spatio attention activity features, and these features are then input to the multi-layer GRU, which learns the human activity frame sequences overall. The proposed model obtains 98.50 % recognition accuracy on the UCF101 dataset and 79.21 % recognition accuracy on the HMDB51 dataset, which is considerably better than the existing models.</div></div>","PeriodicalId":49521,"journal":{"name":"Signal Processing-Image Communication","volume":"137 ","pages":"Article 117308"},"PeriodicalIF":3.4,"publicationDate":"2025-03-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143826393","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Domain-guided multi-frequency underwater image enhancement network","authors":"Qingzheng Wang,&nbsp;Bin Li,&nbsp;Ge Shi,&nbsp;Xinyu Wang,&nbsp;Yiliang Chen","doi":"10.1016/j.image.2025.117281","DOIUrl":"10.1016/j.image.2025.117281","url":null,"abstract":"<div><div>The distribution of underwater images exhibits diverse due to the varied scattering and absorption of light in different water types. However, most existing methods have significant limitations as they cannot distinguish the difference between different water types during enhancement processing, and do not propose clear solutions for the different frequency information. Therefore, the key challenge is to achieve consistency between learned features and water types while preserving multi-frequency information. Thus, we propose a domain-guided multi-frequency underwater image enhancement network (DGMF), which generate high quality images by learning water-type-related features and capturing multi-frequency information. Specifically, we introduce a domain-aware module equipped with a water type classifier, which can distinguish the impacts of different water types, and guide the update of the model towards the specific domain. In addition, we design a multi-frequency mixer that couples Multi-Group Convolution (MGC) and Global Sparse Attention (GSA) to more effectively captures local and global information. Extensive experiments demonstrate that our method outperforms most state-of-the-art methods in both visual perception and evaluation metrics. The code is publicly available at <span><span>https://github.com/liyoucai699/DGMF.git</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49521,"journal":{"name":"Signal Processing-Image Communication","volume":"136 ","pages":"Article 117281"},"PeriodicalIF":3.4,"publicationDate":"2025-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143686524","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"DMCTDet: A density map-guided composite transformer network for object detection of UAV images","authors":"Junjie Li ,&nbsp;Si Guo ,&nbsp;Shi Yi ,&nbsp;Runhua He ,&nbsp;Yong Jia","doi":"10.1016/j.image.2025.117284","DOIUrl":"10.1016/j.image.2025.117284","url":null,"abstract":"<div><div>The application of unmanned aerial vehicles (UAVs) in urban scene object detection is a vital area of research in urban planning, intelligent monitoring, disaster prevention, and urban surveillance.e However, detecting objects in urban scenes captured by UAVs is a challenging task mainly due to the small size of the objects, the variability within the same class, and the diversity of objects. To design an object detection network that can be applied to complex urban scenes, this study proposes a novel composite transformer object detection network guided by a density map (DMCTDet) for urban scene detection in UAV images. The distributional a priori information of objects can be fully exploited by density maps. In the detection stage, a composite backbone feature extraction network is constructed by Swin Transformer combined with Vision Longformer, which can fully extract the scale-variation objects. Adaptive multiscale feature pyramid enhancement modules (AMFPEM) are inserted in the feature fusion stage between both Swin Transformer and Vision Longformer to learn the relationship between object scale variation and enhance the feature representation capacity of small objects. In this way, the accuracy of urban scene detection is significantly improved, and weak aggregated objects are successfully detected from UAV images. Extensive ablation experiments and comparison experiments. of the proposed network are conducted on publicly available urban scene detection datasets of UAV images. The experimental results demonstrate the effectiveness of the designed network structure and the superiority of the proposed network compared to state-of-the-art methods in terms of detection accuracy.</div></div>","PeriodicalId":49521,"journal":{"name":"Signal Processing-Image Communication","volume":"136 ","pages":"Article 117284"},"PeriodicalIF":3.4,"publicationDate":"2025-03-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143817208","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Towards human society-inspired decentralized DNN inference","authors":"Dimitrios Papaioannou,&nbsp;Vasileios Mygdalis,&nbsp;Ioannis Pitas","doi":"10.1016/j.image.2025.117306","DOIUrl":"10.1016/j.image.2025.117306","url":null,"abstract":"<div><div>In human societies, individuals make their own decisions and they may select if and who may influence it, by e.g., consulting with people of their acquaintance or experts of a field. At a societal level, the overall knowledge is preserved and enhanced by individual person empowerment, where complicated consensus protocols have been developed over time in the form of societal mechanisms to assess, weight, combine and isolate individual people opinions. In distributed machine learning environments however, individual AI agents are merely part of a system where decisions are made in a centralized and aggregated fashion or require a fixed network topology, a practice prone to security risks and collaboration is nearly absent. For instance, Byzantine Failures may tamper both the training and inference stage of individual AI agents, leading to significantly reduced overall system performance. Inspired by societal practices, we propose a decentralized inference strategy where each individual agent is empowered to make their own decisions, by exchanging and aggregating information with other agents in their network. To this end, a “Quality of Inference” consensus protocol (QoI) is proposed, forming a single commonly accepted inference rule applied by every individual agent. The overall system knowledge and decisions on specific manners can thereby be stored by all individual agents in a decentralized fashion, employing e.g., blockchain technology. Our experiments in classification tasks indicate that the proposed approach forms a secure decentralized inference framework, that prevents adversaries at tampering the overall process and achieves comparable performance with centralized decision aggregation methods.</div></div>","PeriodicalId":49521,"journal":{"name":"Signal Processing-Image Communication","volume":"136 ","pages":"Article 117306"},"PeriodicalIF":3.4,"publicationDate":"2025-03-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143643817","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Adaptive spatially regularized target attribute-aware background suppressed deep correlation filter for object tracking","authors":"Sathiyamoorthi Arthanari,&nbsp;Sathishkumar Moorthy,&nbsp;Jae Hoon Jeong,&nbsp;Young Hoon Joo","doi":"10.1016/j.image.2025.117305","DOIUrl":"10.1016/j.image.2025.117305","url":null,"abstract":"<div><div>In recent years, deep feature-based correlation filters have attained impressive performance in robust object tracking. However, deep feature-based correlation filters are affected by undesired boundary effects, which reduce the tracking performance. Moreover, the tracker moves towards a region that is identical to the target due to the sudden variation in target appearance and complicated background areas. To overcome these issues, we propose an adaptive spatially regularized target attribute-aware background suppressed deep correlation filter (ASTABSCF). To do this, a novel adaptive spatially regularized technique is presented, which aims to learn an efficient spatial weight for a particular object and fast target appearance variations. Specifically, we present a target-aware background suppression method with dual regression approach, which utilizes a saliency detection technique to produce the target mask. In this technique, we employ the global and target features to get the dual filters known as the global and target filters. Accordingly, global and target response maps are produced by dual filters, which are integrated into the detection stage to optimize the target response. In addition, a novel adaptive attribute-aware approach is presented to emphasize channel-specific discriminative features, which implements a post-processing technique on the observed spatial patterns to reduce the influence of less prominent channels. Therefore, the learned adaptive spatial attention patterns significantly reduce the irrelevant information of multi-channel features and improve the tracker performance. Finally, we demonstrate the efficiency of the ASTABSCF approach against existing modern trackers using the OTB-2013, OTB-2015, TempleColor-128, UAV-123, LaSOT, and GOT-10K benchmark datasets.</div></div>","PeriodicalId":49521,"journal":{"name":"Signal Processing-Image Communication","volume":"136 ","pages":"Article 117305"},"PeriodicalIF":3.4,"publicationDate":"2025-03-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143686523","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Lost in light field compression: Understanding the unseen pitfalls in computer vision","authors":"Adam Zizien ,&nbsp;Chiara Galdi ,&nbsp;Karel Fliegel ,&nbsp;Jean-Luc Dugelay","doi":"10.1016/j.image.2025.117304","DOIUrl":"10.1016/j.image.2025.117304","url":null,"abstract":"<div><div>Could we be overlooking a fundamental aspect of light fields in our quest for efficient compression? The vast amount of data enclosed in a light field makes compression a necessity. Yet, from an application point of view, the focus is predominantly on visual consumption while light fields have properties that can potentially be used in various other tasks. This paper examines the impact of light field compression on the performance of subsequent computer vision tasks. We investigate the variations in quality across perspectives and their impact on face recognition systems and disparity estimation. By leveraging a diverse dataset of light field images, we thoroughly evaluate the performance of various face recognition algorithms when subjected to different conventional and learning-based compression techniques, such as JPEG Pleno, ALVC, and SADN-QVRF. Our findings reveal a noticeable decline in peak recognition performance as compression levels increase, given specific recognition frameworks. Furthermore, we identify a significant shift in the recognition threshold, particularly in response to higher degrees of compression. Secondly, by relying on a novel disparity estimation algorithm, we explore the loss of information across light field perspectives. Our results highlight a disconnect between the preservation of visual fidelity and the loss of minute detail crucial for the preservation of disparity information in light field images. The findings presented herein aim to contribute to the development of efficient compression strategies while emphasizing the delicate balance between compression efficiency, subjective quality, and feature preservation with the aim of increased accuracy in specialized light field systems.</div></div>","PeriodicalId":49521,"journal":{"name":"Signal Processing-Image Communication","volume":"136 ","pages":"Article 117304"},"PeriodicalIF":3.4,"publicationDate":"2025-03-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143620241","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Underwater image quality assessment method via the fusion of visual and structural information","authors":"Tianhai Chen ,&nbsp;Xichen Yang ,&nbsp;Tianshu Wang ,&nbsp;Nengxin Li ,&nbsp;Shun Zhu ,&nbsp;Genlin Ji","doi":"10.1016/j.image.2025.117285","DOIUrl":"10.1016/j.image.2025.117285","url":null,"abstract":"<div><div>Underwater-captured images often suffer from quality degradation due to the challenging underwater environment, leading to information loss that significantly affects their usability. Therefore, accurately predicting the quality of underwater images is crucial. To tackle this issue, this study introduces a novel Underwater Image Quality Assessment method that combines visual and structural information. First, the CIELab map, gradient feature map, and Mean Subtracted Contrast Normalized feature map of the underwater image are obtained. Then, these feature maps are divided into non-overlapping 32x32 patches, and each patch is fed into the corresponding sub-network. This method allows for a comprehensive description of the changes in visual and structural information resulting from quality degradation in underwater images. Subsequently, the features extracted by the multipath network are fused using a feature fusion network to promote feature complementarity and overcome the limitations of individual features. Finally, the relationship between underwater image quality and fusion features was learned to obtain an evaluation model. Furthermore, the quality of the underwater image can be measured by combining the quality prediction scores of different patches. Experimental results on underwater image datasets demonstrate that the proposed method can achieve more accurate and stable quality measurement results with a more lightweight structure. Meanwhile, performance comparisons on natural image datasets and screen content image datasets confirm that the proposed method is more applicable for complex application scenarios than existing methods. The code is open-source and available at <span><span>https://github.com/dart-into/UIQAVSI</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49521,"journal":{"name":"Signal Processing-Image Communication","volume":"136 ","pages":"Article 117285"},"PeriodicalIF":3.4,"publicationDate":"2025-02-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143527063","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Advanced transformer for high-noise image denoising: Enhanced attention and detail preservation","authors":"Jie Zhang ,&nbsp;Wenxiao Huang ,&nbsp;Miaoxin Lu ,&nbsp;Fengxian Wang ,&nbsp;Mingdong Zhao ,&nbsp;Yinhua Li","doi":"10.1016/j.image.2025.117286","DOIUrl":"10.1016/j.image.2025.117286","url":null,"abstract":"<div><div>In image denoising, the transformer model effectively captures global dependencies within an image due to its self-attention mechanism. This capability enhances the understanding of the overall structure and details of the image during the denoising process. However, the computational complexity of global self-attention increases quadratically with higher spatial resolutions, making it unsuitable for the real-time denoising of high-resolution and high-noise images. And, the use of local windows alone neglects the long-range pixel correlations. Furthermore, the self-attention mechanism applies a global weighting to the pixels of the input image, which can lead to the smoothing or loss of fine details. To enrich structural information and alleviate the computational complexity associated with global self-attention, we propose an edge-enhanced windowed multi-head self-attention mechanism (EWMSA). This mechanism combines edge enhancement with windowed self-attention to reduce computational demands while allowing edge features to better preserve detail and texture information. To mitigate the effects of ineffective features with low weights, we introduce a feed-forward network with a gate control strategy (LGFN). This network adjusts pixel weights to prioritize attention on effective pixels, thereby enhancing their prominence. Furthermore, to compensate for the limitations of window-based self-attention in global pixel utilization, we propose a deformable convolution block (DFCB). This block improves the interaction of contextual information and allows for better adaptation to texture variations within the image. Extensive experiments demonstrate that the proposed ATHID is competitive with other state-of-the-art denoising methods when applied to real-world noise and various synthetic noise levels, effectively addressing the challenges of high-noise image denoising. The code and models are publicly available at <span><span>https://github.com/zzuli407/ATHID</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49521,"journal":{"name":"Signal Processing-Image Communication","volume":"136 ","pages":"Article 117286"},"PeriodicalIF":3.4,"publicationDate":"2025-02-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143550934","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Visually multimodal depression assessment based on key questions with weighted multi-task learning","authors":"Peng Wang ,&nbsp;Miaomiao Cao ,&nbsp;Xianlin Zhu ,&nbsp;Suhong Wang ,&nbsp;Rongrong Ni ,&nbsp;Changchun Yang ,&nbsp;Biao Yang","doi":"10.1016/j.image.2025.117279","DOIUrl":"10.1016/j.image.2025.117279","url":null,"abstract":"<div><div>In recent years, depression has received attention due to its high prevalence and high risk of suicide. In contrast, the increased pressure on health care and the shortage of mental health professionals have led to the failure to detect and intervene in depression promptly. To solve the above problems, we propose a visual multi-modal fusion network for depression assessment based on weighted multi-task learning (WMTL). First, the visual cues of different modalities are collected from the subjects when they answer key questions in the simulated interview to mitigate redundancy. Afterward, spatial attention-based feature embedding modules are proposed to extract depression-aware features from different visual cues. Finally, a hierarchical weighted attention fusion (HAF) module is presented to fuse the depression-aware features from different modalities and facilitate depression assessment. Comprehensive evaluations are conducted on the benchmarking DAIC-WOZ. Experimental results show that the proposed method performs well in assessing depression, with an average accuracy of 76.96% for ten questions and an F1 score of 0.85. The high performance also indicates a strong correlation between key questions in the interview and depression levels.</div></div>","PeriodicalId":49521,"journal":{"name":"Signal Processing-Image Communication","volume":"135 ","pages":"Article 117279"},"PeriodicalIF":3.4,"publicationDate":"2025-02-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143445380","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Exploiting rank-based filter pruning for real-time UAV tracking","authors":"Xucheng Wang ,&nbsp;Dan Zeng ,&nbsp;Qijun Zhao ,&nbsp;Shuiwang Li","doi":"10.1016/j.image.2025.117278","DOIUrl":"10.1016/j.image.2025.117278","url":null,"abstract":"<div><div>UAV tracking is an emerging task and has wide potential applications in such as agriculture, navigation, entertainment and public security. However, the limitations of computing resources, battery capacity, and maximum load of UAV hinder the deployment of DL-based tracking algorithms on UAV. In contrast to deep learning trackers, discriminative correlation filters (DCF)-based trackers stand out in the UAV tracking community because of their high efficiency. However, their precision is usually much lower than trackers based on deep learning. Model compression is a promising way to reduce the disparity (i.e., efficiency, precision) between DCF- and deep learning- based trackers, which has not caught much attention in the UAV tracking community. In this paper, We propose the P-SiamFC++ tracker, which is the first to use rank-based filter pruning to compress the SiamFC++ model, achieving a remarkable balance between efficiency and precision. Our method is general and could inspire additional research into UAV tracking with model compression in the future. Extensive experiments on four UAV benchmarks, including UAV123@10fps, DTB70, UAVDT and Vistrone2018, show that P-SiamFC++ tracker significantly outperforms state-of-the-art UAV tracking methods.</div></div>","PeriodicalId":49521,"journal":{"name":"Signal Processing-Image Communication","volume":"135 ","pages":"Article 117278"},"PeriodicalIF":3.4,"publicationDate":"2025-02-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143464855","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}