Signal Processing-Image Communication最新文献

{"title":"Lp-norm distortion-efficient adversarial attack","authors":"Chao Zhou , Yuan-Gen Wang , Zi-Jia Wang , Xiangui Kang","doi":"10.1016/j.image.2024.117241","DOIUrl":"10.1016/j.image.2024.117241","url":null,"abstract":"<div><div>Adversarial examples have shown a powerful ability to make a well-trained model misclassified. Current mainstream adversarial attack methods only consider one of the distortions among <span><math><msub><mrow><mi>L</mi></mrow><mrow><mn>0</mn></mrow></msub></math></span>-norm, <span><math><msub><mrow><mi>L</mi></mrow><mrow><mn>2</mn></mrow></msub></math></span>-norm, and <span><math><msub><mrow><mi>L</mi></mrow><mrow><mi>∞</mi></mrow></msub></math></span>-norm. <span><math><msub><mrow><mi>L</mi></mrow><mrow><mn>0</mn></mrow></msub></math></span>-norm based methods cause large modification on a single pixel, resulting in naked-eye visible detection, while <span><math><msub><mrow><mi>L</mi></mrow><mrow><mn>2</mn></mrow></msub></math></span>-norm and <span><math><msub><mrow><mi>L</mi></mrow><mrow><mi>∞</mi></mrow></msub></math></span>-norm based methods suffer from weak robustness against adversarial defense since they always diffuse tiny perturbations to all pixels. A more realistic adversarial perturbation should be sparse and imperceptible. In this paper, we propose a novel <span><math><msub><mrow><mi>L</mi></mrow><mrow><mi>p</mi></mrow></msub></math></span>-norm distortion-efficient adversarial attack, which not only owns the least <span><math><msub><mrow><mi>L</mi></mrow><mrow><mn>2</mn></mrow></msub></math></span>-norm loss but also significantly reduces the <span><math><msub><mrow><mi>L</mi></mrow><mrow><mn>0</mn></mrow></msub></math></span>-norm distortion. To this aim, we design a new optimization scheme, which first optimizes an initial adversarial perturbation under <span><math><msub><mrow><mi>L</mi></mrow><mrow><mn>2</mn></mrow></msub></math></span>-norm constraint, and then constructs a dimension unimportance matrix for the initial perturbation. Such a dimension unimportance matrix can indicate the adversarial unimportance of each dimension of the initial perturbation. Furthermore, we introduce a new concept of adversarial threshold for the dimension unimportance matrix. The dimensions of the initial perturbation whose unimportance is higher than the threshold will be all set to zero, greatly decreasing the <span><math><msub><mrow><mi>L</mi></mrow><mrow><mn>0</mn></mrow></msub></math></span>-norm distortion. Experimental results on three benchmark datasets show that under the same query budget, the adversarial examples generated by our method have lower <span><math><msub><mrow><mi>L</mi></mrow><mrow><mn>0</mn></mrow></msub></math></span>-norm and <span><math><msub><mrow><mi>L</mi></mrow><mrow><mn>2</mn></mrow></msub></math></span>-norm distortion than the state-of-the-art. Especially for the MNIST dataset, our attack reduces 8.1% <span><math><msub><mrow><mi>L</mi></mrow><mrow><mn>2</mn></mrow></msub></math></span>-norm distortion meanwhile remaining 47% pixels unattacked. This demonstrates the superiority of the proposed method over its competitors in terms of adversarial robustness and visual imperceptibility. The code is avail","PeriodicalId":49521,"journal":{"name":"Signal Processing-Image Communication","volume":"131 ","pages":"Article 117241"},"PeriodicalIF":3.4,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143129355","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":"Data-driven gradient priors integrated into blind image deblurring","authors":"Qing Qi ,&nbsp;Jichang Guo ,&nbsp;Chongyi Li","doi":"10.1016/j.image.2025.117275","DOIUrl":"10.1016/j.image.2025.117275","url":null,"abstract":"<div><div>Blind image deblurring is a severely ill-posed task. Most existing methods focus on deep learning to learn massive data features while ignoring the vital significance of classic image structure priors. We make extensive use of the image gradient information in a data-driven way. In this paper, we present a Generative Adversarial Network (GAN) architecture based on image structure priors for blind non-uniform image deblurring. Previous image deblurring methods employ Convolutional Neural Networks (CNNs) and non-blind deconvolution algorithms to predict kernel estimations and obtain deblurred images, respectively. We permeate the structure prior of images throughout the design of network architectures and target loss functions. To facilitate network optimization, we propose multi-term target loss functions aimed to supervise the generator to have images with significant structure attributes. In addition, we design a dual-discriminant mechanism for discriminating whether the image edge is clear or not. Not only image content but also the sharpness of image structures need to be discriminated. To learn image gradient features, we develop a dual-flow network that considers both the image and gradient domains to learn image gradient features. Our model directly avoids the accumulated errors caused by two steps of “kernel estimation-non-blind deconvolution”. Extensive experiments on both synthetic datasets and real-world images demonstrate that our model outperforms state-of-the-art methods.</div></div>","PeriodicalId":49521,"journal":{"name":"Signal Processing-Image Communication","volume":"135 ","pages":"Article 117275"},"PeriodicalIF":3.4,"publicationDate":"2025-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143350862","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":"DarkSegNet: Low-light semantic segmentation network based on image pyramid","authors":"Jintao Tan,&nbsp;Longyang Huang,&nbsp;Zhonghui Chen,&nbsp;Ruokun Qu,&nbsp;Chenglong Li","doi":"10.1016/j.image.2025.117265","DOIUrl":"10.1016/j.image.2025.117265","url":null,"abstract":"<div><div>In the domain of computer vision, the task of semantic segmentation for images captured under low-light conditions has proven to be a formidable challenge. To address this challenge, we introduce a novel low-light semantic segmentation model named DarkSegNet. The DarkSegNet model aims to deal with the problem of semantic segmentation of low-light images. It effectively mines potential information in images by combining image pyramid decomposition, spatial low-frequency attention (SLA) module, and channel low-frequency information enhancement (CLIE) module to achieve better low-light semantic segmentation performance. These components work synergistically to effectively extract latent information embedded within the low-light image, ultimately resulting in improved performance of low-light semantic segmentation. We conduct experiments on the UAV indoor low-light LLRGBD-real dataset. Compared to other mainstream semantic segmentation methods, DarkSegNet achieves the highest mIoU of 47.9% on the UAV indoor low-light LLRGBD-real dataset. It is worth emphasizing that our model implements end-to-end training, avoiding the need to design additional image enhancement modules. The DarkSegNet network holds significant potential for facilitating drone-based rescue operations in disaster-stricken environments.</div></div>","PeriodicalId":49521,"journal":{"name":"Signal Processing-Image Communication","volume":"135 ","pages":"Article 117265"},"PeriodicalIF":3.4,"publicationDate":"2025-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143157474","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":"UIEVUS: An underwater image enhancement method for various underwater scenes","authors":"Siyi Ren ,&nbsp;Xianqiang Bao ,&nbsp;Tianjiang Wang ,&nbsp;Xinghua Xu ,&nbsp;Tao Ma ,&nbsp;Kun Yu","doi":"10.1016/j.image.2025.117264","DOIUrl":"10.1016/j.image.2025.117264","url":null,"abstract":"<div><div>Due to the scattering and absorption of light in water, underwater images commonly encounter degradation issues, such as color distortions and uneven brightness. To address these challenges, we introduce UIEVUS, an underwater image enhancement method designed for various underwater scenes. Building upon Retinex theory, our method implements an approach that combines Retinex decomposition with generative adversarial learning for targeted enhancement. The core innovation of UIEVUS lies in its ability to separately process and recover illumination and reflection maps before merging them into the final enhanced result. Specifically, the method first applies Retinex decomposition to separate the original underwater image into an illumination map (addressing uneven lighting) and a reflection map (addressing color distortion). The reflection map undergoes restoration through a lightweight encoder–decoder network that employs generative adversarial learning to recover color information. Concurrently, the illumination map receives enhancement guided by the reflection map, resulting in improved edges, details, brightness, and reduced noise. These enhanced components are then merged to produce the final result. Extensive experimental results demonstrate that UIEVUS achieves competitive performance against other comparative algorithms across various benchmark tests. Notably, our method strikes an optimal balance between computational efficiency and enhancement quality, making it suitable for practical UUV applications.</div></div>","PeriodicalId":49521,"journal":{"name":"Signal Processing-Image Communication","volume":"135 ","pages":"Article 117264"},"PeriodicalIF":3.4,"publicationDate":"2025-01-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143157475","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":"An Efficient Dehazing Method Using Pixel Unshuffle and Color Correction","authors":"Hongyuan Jing ,&nbsp;Kaiyan Wang ,&nbsp;Zhiwei Zhu ,&nbsp;Aidong Chen ,&nbsp;Chen Hong ,&nbsp;Mengmeng Zhang","doi":"10.1016/j.image.2025.117260","DOIUrl":"10.1016/j.image.2025.117260","url":null,"abstract":"<div><div>Severe weather conditions such as haze and rainstorm will lead to serious degradation of observed images, which will influence the performance of advanced visual tasks such as target detection. However, most of the existing image processing methods focus on dehazing while overlooking the restoration of image color and details. In this paper, we found that the variance of the RGB three channels of a pixel at a certain point in an RGB image is related to its corresponding degree of color brightness through a large number of experiments, and propose an efficient dehazing method called PUCCNet, which utilizes Pixel Unshuffle and Color Correction to enhance image detail information and improve color saturation. We designed a Detail Recover Block (DRB) in the network to capture the details of the input image and focus on local details through the attention mechanism. In the high-dimensional part of the network, a Depth Local Global Residual Block (DLGRB) is introduced, which can simultaneously handle local and global features, thereby enhancing the model's expressive capability, improving its generalization ability, and reducing the risk of overfitting. The network obtains local details through the attention mechanism, and makes the output image of higher quality through color correction, which is aligned with the human visual system. Extensive experiments on synthetic datasets and real-world datasets demonstrate that the proposed method outperforms existing state-of-the-art methods.</div></div>","PeriodicalId":49521,"journal":{"name":"Signal Processing-Image Communication","volume":"134 ","pages":"Article 117260"},"PeriodicalIF":3.4,"publicationDate":"2025-01-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143171432","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":"Scene recovery with detail-preserving","authors":"Tingting Wu ,&nbsp;Xinru Wang ,&nbsp;Jun Liu ,&nbsp;Tieyong Zeng","doi":"10.1016/j.image.2025.117266","DOIUrl":"10.1016/j.image.2025.117266","url":null,"abstract":"<div><div>Images captured in sandstorms, hazy, snowy or underwater conditions often suffer from poor visibility. This is mainly due to the presence of atmospheric particles that scatter light. Based on the assumption of highly linear correlation between <span><math><msub><mrow><mi>S</mi></mrow><mrow><mi>n</mi><mi>u</mi></mrow></msub></math></span> and the observed intensity <span><math><mi>I</mi></math></span>, we first estimate the scattering map <span><math><mover><mrow><mi>t</mi></mrow><mrow><mo>̃</mo></mrow></mover></math></span> by projecting the input image <span><math><mi>I</mi></math></span> onto the unified spectrum <span><math><msub><mrow><mi>S</mi></mrow><mrow><mi>n</mi><mi>u</mi></mrow></msub></math></span>. We then apply the weighted guided image filter to make the corresponding transmission map <span><math><mi>t</mi></math></span> more accurate so that details and textures of the input image can be better recovered. Since the atmospheric light <span><math><mi>A</mi></math></span> is also critical to the scene recovery, we propose to use the quad-tree subdivision to extract a correct <span><math><mi>A</mi></math></span>. The quantitative and qualitative evaluations are reported in the numerical experiments. Compared with some SOTA methods, the images recovered by our method exhibit better visibility while preserving details.</div></div>","PeriodicalId":49521,"journal":{"name":"Signal Processing-Image Communication","volume":"134 ","pages":"Article 117266"},"PeriodicalIF":3.4,"publicationDate":"2025-01-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143171427","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":"PointPCA+: A full-reference Point Cloud Quality Assessment metric with PCA-based features","authors":"Xuemei Zhou ,&nbsp;Evangelos Alexiou ,&nbsp;Irene Viola ,&nbsp;Pablo Cesar","doi":"10.1016/j.image.2025.117262","DOIUrl":"10.1016/j.image.2025.117262","url":null,"abstract":"<div><div>This paper introduces an enhanced Point Cloud Quality Assessment (PCQA) metric, termed PointPCA+, as an extension of PointPCA, with a focus on computational simplicity and feature richness. PointPCA+ refines the original PCA-based descriptors by employing Principal Component Analysis (PCA) solely on geometry data; additionally, the texture descriptors are refined through a direct application of the function on YCbCr values, enhancing the efficiency of computation. The metric combines geometry and texture features, capturing local shape and appearance properties, through a learning-based fusion to generate a total quality score. Prior to fusion, a feature selection module is incorporated to identify the most effective features from a proposed super-set. Experimental results demonstrate the high predictive performance of PointPCA+ against subjective ground truth scores obtained from four publicly available datasets. The metric consistently outperforms state-of-the-art solutions, offering valuable insights into the design of similarity measurements and the effectiveness of handcrafted features across various distortion types. The code of the proposed metric is available at <span><span>https://github.com/cwi-dis/pointpca_suite/</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49521,"journal":{"name":"Signal Processing-Image Communication","volume":"135 ","pages":"Article 117262"},"PeriodicalIF":3.4,"publicationDate":"2025-01-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143157473","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":"Optimizing multi-task network with learned prototypes for weakly supervised semantic segmentation","authors":"Lei Zhou ,&nbsp;Jiasong Wang ,&nbsp;Jing Luo ,&nbsp;Yuheng Guo ,&nbsp;Xiaoxiao Li","doi":"10.1016/j.image.2025.117272","DOIUrl":"10.1016/j.image.2025.117272","url":null,"abstract":"<div><div>Weakly supervised semantic segmentation (WSSS) presents a challenging task wherein semantic objects are extracted solely through the utilization of image-level labels as supervision. One common category of state-of-the-art solutions depends on the generation of pseudo pixel-level annotations via the use of localization maps. Nevertheless, in the majority of such solutions, the quality of pseudo annotations may not effectively fulfill the requirements of semantic segmentation owing to the incomplete nature of the localization maps. In order to generate denser localization maps for WSSS, this paper proposes the use of a prototype learning guided multi-task network. Initially, the prototypes (also referred to as prototypical feature vectors) are employed to depict the similarities between images. Specifically, the shared information among different training images is thoroughly exploited to concomitantly learn the prototypes for both foreground categories and background. This approach facilitates the localization of more reliable background pixels and foreground regions by evaluating the similarities between the representative prototypes and the extracted features of pixels. Additionally, the learned prototypes can be incorporated into the multi-task network to enhance the efficiency of parameter optimization by adaptively rectifying errors in pixel-level supervision. Therefore, the optimization of the multi-task network for object localization and the production of high-quality proxy annotations can be achieved by means of clean image-level labels and refined pixel-level supervision working in conjunction. By selecting and refining proxy annotations, the performance of the segmentation algorithm can be further improved. Extensive experiments conducted on two datasets, namely, PASCAL VOC 2012 and COCO 2014, have substantiated the fact that the prototype learning guided multi-task network being proposed outperforms the current state-of-the-art (SOTA) methods in terms of segmentation performance, achieving a mean IoU of 72.1% and 72.6% on the PASCAL VOC 2012 validation and test sets, respectively.</div></div>","PeriodicalId":49521,"journal":{"name":"Signal Processing-Image Communication","volume":"134 ","pages":"Article 117272"},"PeriodicalIF":3.4,"publicationDate":"2025-01-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143171426","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":"Physics prior-based contrastive learning for low-light image enhancement","authors":"Hongxiang Liu,&nbsp;Yunliang Zhuang,&nbsp;Chen Lyu","doi":"10.1016/j.image.2025.117274","DOIUrl":"10.1016/j.image.2025.117274","url":null,"abstract":"<div><div>Capturing images in low-light conditions can lead to losing image content, making low-light image enhancement a practically challenging task. Various deep-learning methods have been proposed to address this challenge, demonstrating significant progress. However, existing methods still face challenges in achieving uniform brightness enhancement. These methods rely solely on normal-light images to guide the training of the enhancement network, resulting in insufficient utilization of low-light image information. We propose a novel Illumination Contrastive Learning (ICL) that employs positive and negative samples to improve contrast relationships and combines local brightness data to align luminance images closer to normal light and away from low-light areas. Existing methods that use channel attention mechanisms often neglect global channel dependencies, leading to poor color contrast in enhanced images. We address this issue by developing a Multi-scale Channel Dependency Representation Block (MCRB) that utilizes multi-scale attention to capture a wide range of channel dependencies, thereby enhancing contrast more effectively. Based on the Retinex theory, our method maximizes the use of illumination information in low-light images and integrates contrast learning into a Retinex-based framework. This integration results in a more uniform brightness distribution and improved visual effects in enhanced images. The effectiveness of our method has been validated through tests on various synthetic and natural datasets, surpassing existing state-of-the-art methods.</div></div>","PeriodicalId":49521,"journal":{"name":"Signal Processing-Image Communication","volume":"134 ","pages":"Article 117274"},"PeriodicalIF":3.4,"publicationDate":"2025-01-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143171429","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":"Anti-noise face: A resilient model for face recognition with labeled noise data","authors":"Lei Wang,&nbsp;Xun Gong,&nbsp;Jie Zhang,&nbsp;Rui Chen","doi":"10.1016/j.image.2025.117269","DOIUrl":"10.1016/j.image.2025.117269","url":null,"abstract":"<div><div>With the remarkable success of face recognition driven by large-scale datasets, noise learning has gained increasing attention due to the prevalence of noise within these datasets. While various margin-based loss functions and training strategies for label noise have been recently devised, two issues still remain to consider: (1) The explicit emphasis on specific characteristics of different types of noise is required. (2) The potential impact of noise during the early stages of training, which may lead to convergence issues, should not be ignored. In this study, we propose a comprehensive algorithm for learning with label noise. Compared to the existing noise self-correction methods, we further enhance detecting closed-set noise by introduce a closed-set noise self-correction module, and introduce a novel loss function for handling remaining noisy samples detected by an improved Gaussian Mixture Model. Additionally, we use a progressive approach, where we work through the easy examples first and then move on to the difficult ones, just as a student work through a course with the easy ones first and then the difficult ones later. Extensive experiments conducted on the synthesized noise dataset and on popular benchmarks have demonstrated the superior effectiveness of our approach over state-of-the-art alternatives.</div></div>","PeriodicalId":49521,"journal":{"name":"Signal Processing-Image Communication","volume":"134 ","pages":"Article 117269"},"PeriodicalIF":3.4,"publicationDate":"2025-01-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143172443","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}