Search Results (173)

Skin lesion segmentation has achieved notable progress in recent years; however, accurate delineation remains challenging due to complex backgrounds, ambiguous boundaries, and low lesion-to-skin contrast. To address these issues, we propose the lesion-aware triple-path feature fusion network (LTPNet), an end-to-end framework that progressively processes features through extraction, refinement, and aggregation stages. In the extraction stage, we incorporate a general foreground–background attention to suppress background interference and accelerate model convergence. In the refinement stage, we introduce an attentive spatial modulator (ASM) to jointly exploit local structural cues and global semantic context for precise spatial modulation. We further develop a lesion-aware lite-gate attention (LALGA) module that performs local spatial feature modulation and global channel recalibration tailored to lesion characteristics. In the aggregation stage, we propose a triple-path feature fusion (TPFF) module that explicitly models feature relationships across scales via three complementary pathways: a common path (CP) for semantic consistency, a saliency path (SP) for highlighting co-activated regions, and a difference path (DP) for accentuating structural discrepancies. Extensive experiments on in-domain and cross-domain datasets show that LTPNet achieves superior segmentation accuracy with reasonable inference efficiency and model complexity, demonstrating its potential for efficient and reliable clinical decision support. Full article

Semi-supervised medical image segmentation has recently achieved great success, but assigning trustworthy pseudo-labels to unlabeled images has been a difficult problem in medical image processing. A common solution is to select reliable predicted pixels as the pseudo-labels. However, unreliable pixels are often concentrated in the edge areas of the foreground and background in medical tasks. Directly discarding these pixels will result in this important information never being available. The foreground of medical images is usually surrounded by the edge area. This section of pixels is a mixture of the two categories, which makes it very difficult to distinguish. To address these problems, we propose a semi-supervised medical segmentation framework that combines conformal prediction and contrastive learning. Our framework can use conformal prediction to select pseudo-labels with high confidence and preserve important boundary information. Furthermore, the segmentation performance of edge regions can be improved using contrastive learning between edge categories and non-edge categories. Extensive experiments on multiple benchmarks show that our framework consistently outperforms state-of-the-art methods. Full article

Remote sensing-based landslide segmentation is of great significance for geological hazard assessment and post-disaster rescue. Existing convolutional neural network methods, constrained by the inherent limitations of spatial convolution, tend to lose high-frequency edge details during deep semantic extraction, while frequency-domain analysis, although capable of globally preserving high-frequency components, struggles to perceive local multi-scale features. The lack of an effective synergistic mechanism between them makes it difficult for networks to balance regional integrity and boundary precision. To address these issues, this paper proposes the Geometric Context and Frequency Domain Fusion Network (GCF-Net), which achieves explicit edge enhancement through a three-stage progressive framework. First, the Pyramid Lightweight Fusion (PGF) block is proposed to aggregate multi-scale context and provide rich hierarchical features for subsequent stages. Second, the Geometric Context and Frequency Domain Fusion (GCF) module is designed, where the frequency-domain branch generates dynamic high-frequency masks via the Fourier transform to locate boundary positions, while the spatial branch models foreground–background relationships to understand boundary semantics, with both branches fused through an adaptive gating mechanism. Finally, Edge-aware Detail Consistency Improvement (EDCI) module is designed to balance boundary preservation and noise suppression based on edge confidence, achieving adaptive output refinement. Under the joint supervision of Focal loss, Dice loss, and Edge loss, experiments on the mixed dataset and LMHLD dataset demonstrate that GCF-Net achieves OAs of 96.42% and 96.71%, respectively. Ablation experiments and visualization results further validate the effectiveness of each module and the significant improvement in boundary segmentation. Full article

Accurate segmentation of the intercondylar notch of the femur is of great significance for the diagnosis of knee joint diseases, surgical planning, and anterior cruciate ligament (ACL) reconstruction. Among them, the obvious anatomical heterogeneity, the interference of structurally similar tissues, and the blurred boundaries in MRI images make the segmentation of the intercondylar notch challenging. The segmentation of the intercondylar notch is often regarded as a standard semantic segmentation problem, but doing so leaves the inherent high-order internal variation and low-contrast features of its anatomical structure unresolved. We proposed a new Wavelet Multi-scale Contextual Attention Network (WMCA-Net). We have coordinated the Shallow High-frequency Feature Dense Extraction Block (SHFDEB) and Wavelet Split and Fusion Block (WSFB) modules with each other. The SHFDEB intensively extracts high-frequency detailed features at the shallowest layer of the network, while the WSFB effectively splits and fuses features at various resolutions, suppressing noise while better preserving the high-frequency detailed structural information we need. The Multi-scale Depth-wise Convolution Block (MDCB) captures cross-scale features from the narrow intercondylar notch (5–8 mm wide) to the surrounding femoral structure (approximately 50 mm diameter), dynamically adapting to different morphologies, including pathological changes caused by osteophyte formation. The Contextual-Weighted Attention Module (CWAM) establishes long-term semantic associations between fuzzy regions and clear anatomical landmarks by precisely locating uncertain regions through foreground and background decomposition. The Dice Similarity Coefficient of WMCA-Net on the intercondylar notch dataset is 93.16%, and the 95% Hausdorff Distance is 1.42 mm, demonstrating its advanced segmentation performance and good anatomical adaptability. Full article

Medical image segmentation is essential for clinical decision-making, treatment planning, and disease monitoring. However, ambiguous boundaries and complex anatomical structures continue to pose challenges for accurate segmentation. To address these issues, we propose DCANet (Disentangled and Category-Aware Network), a novel framework that effectively integrates local and global feature representations while enhancing category-aware feature interactions. In DCANet, features from convolutional and Transformer layers are fused using the Feature Coupling Unit (FCU), which aligns and combines local and global information across multiple semantic levels. The Decoupled Feature Module (DFM) then separates high-level representations into multi-class foreground and background features, improving discriminability and mitigating boundary ambiguity. Finally, the Category-Aware Integration Aggregator (CAIA) guides multi-level feature fusion, emphasizes critical regions, and refines segmentation boundaries. Extensive experiments on four public datasets—Synapse, ACDC, GlaS, and MoNuSeg—demonstrate the superior performance of DCANet, achieving Dice scores of 84.80%, 94.07%, 94.60%, and 79.85%, respectively. These results confirm the effectiveness and generalizability of DCANet in accurately segmenting complex anatomical structures and resolving boundary ambiguities across diverse medical image segmentation tasks. Full article

Ultrasound imaging is widely used for early breast cancer screening to enhance patient survival. However, interpreting these images is inherently challenging due to speckle noise, low lesion-to-tissue contrast, and highly variable tumor morphology within complex anatomical structures. Additionally, variations in image characteristics across institutions and devices further impede the development of robust and generalizable computer-aided diagnostic systems. To alleviate these issues, this paper presents a cross-domain segmentation prior guided classification strategy for robust breast tumor diagnosis in ultrasound imaging, implemented through a novel Dual-Stream Diagnosis Network (DSDNet). DSDNet adopts a decoupled dual-stream architecture, where a frozen segmentation branch supplies spatial priors to guide the classification backbone. This design enables stable and accurate performance across diverse imaging conditions and clinical settings. To realize the proposed DSDNet framework, three novel modules are created. The Dual-Stream Mask Attention (DSMA) module enhances lesion priors by jointly modeling foreground and background cues. The Segmentation Prior Guidance Fusion (SPGF) module integrates multi-scale priors into the classification backbone using cross-domain spatial cues, improving tumor morphology representation. The Mamba-Inspired Linear Transformer (MILT) block, built upon the Mamba-Inspired Linear Attention (MILA) mechanism, serves as an efficient attention-based feature extractor. On the BUSI, BUS, and GDPH_SYSUCC datasets, DSDNet achieves ACC values of 0.878, 0.836, and 0.882, and Recall scores of 0.866, 0.789, and 0.878, respectively. These results highlight the effectiveness and strong classification performance of our method in ultrasound breast cancer diagnosis. Full article

Foreground segmentation and background subtraction are critical components in many computer vision applications, such as intelligent video surveillance, urban security systems, and obstacle detection for autonomous vehicles. Although extensively studied over the past decades, these tasks remain challenging, particularly due to rapid illumination changes, dynamic backgrounds, cast shadows, and camera movements. The emergence of supervised deep learning-based methods has significantly enhanced performance, surpassing traditional approaches on the benchmark dataset CDnet2014. In this context, this paper provides a comprehensive review of recent supervised deep learning techniques applied to background subtraction, along with an in-depth comparative analysis of state-of-the-art approaches available on the official CDnet2014 results platform. Specifically, we examine several key architecture families, including convolutional neural networks (CNN and FCN), encoder–decoder models such as FgSegNet and Motion U-Net, adversarial frameworks (GAN), Transformer-based architectures, and hybrid methods combining intermittent semantic segmentation with rapid detection algorithms such as RT-SBS-v2. Beyond summarizing existing works, this review contributes a structured cross-family comparison under a unified benchmark, a focused analysis of performance behavior across challenging CDnet2014 scenarios, and a critical discussion of the trade-offs between segmentation accuracy, robustness, and computational efficiency for practical deployment. Full article

Street networks shape urban dynamics. However, at the important meso- and micro-scales, a research limitation remains in systematically linking the spatial logic of streets to the physical configuration of street-level commerce, in particular through an analytical lens that distinguishes between different urban network functions. With a view to overcoming this limitation and extending space syntax theory into the fine-grained analysis of commercial form, this study applies its dual-network logic, contrasting foreground networks and background networks. The spatial patterns of street stores were analyzed across eight street segments in four Chinese cities: Tianjin, Nanjing, Zhengzhou, and Hong Kong. Network types were distinguished using Normalized Angular Choice and patchwork pattern analysis. By using 2019 POI data, Street View imagery, and field surveys, a comparative quantitative analysis was conducted across three metrics: operation methods, functional diversity, and 100-m density. The results indicate differences: chain stores hold a clear advantage in high-value segments of the foreground network, a pattern supported by statistical tests. These segments also exhibit higher functional diversity (mean ENT = 5.12). In contrast, high-value street segments of the background network exhibit a consistently higher prevalence of sole stores. They also have a commercial density approximately 2.6 times greater than that of their foreground counterparts. These findings provide empirical evidence on how foreground and background networks support different kinds of commercial ecologies: one oriented toward micro-economy efficiency and standardized supply, the other toward socio-culturally embedded, high-intensity local exchange. Consequently, by linking specific street spatial configurations to measurable commercial outcomes, this research contributes methodologically by operationalizing the dual-network framework at a novel scale and offering a replicable analytical tool for diagnosing and guiding commercial spatial planning in cities. Full article

Recently, deep learning methods have seen major advancements and are preferred for medical image analysis. Clinically, deep learning techniques for cancer image analysis are among the main applications for early diagnosis, detection, and treatment. Consequently, segmentation of breast histology images is a key step towards diagnosing breast cancer. However, the use of deep learning methods for image analysis is constrained by challenging features in the histology images. These challenges include poor image quality, complex microscopic tissue structures, topological intricacies, and boundary/edge inhomogeneity. Furthermore, this leads to a limited number of images required for analysis. The U-Net model was introduced and gained significant traction for its ability to produce high-accuracy results with very few input images. Many modifications of the U-Net architecture exist. Therefore, this study proposes the watershed encoder–decoder neural network (WEDN) to segment cancerous lesions in supervised breast histology images. Pre-processing of supervised breast histology images via augmentation is introduced to increase the dataset size. The augmented dataset is further enhanced and segmented into the region of interest. Data enhancement methods such as thresholding, opening, dilation, and distance transform are used to highlight foreground and background pixels while removing unwanted parts from the image. Consequently, further segmentation via the connected component analysis method is used to combine image pixel components with similar intensity values and assign them their respective labeled binary masks. The watershed filling method is then applied to these labeled binary mask components to separate and identify the edges/boundaries of the regions of interest (cancerous lesions). This resultant image information is sent to the WEDN model network for feature extraction and learning via training and testing. Residual convolutional block layers of the WEDN model are the learnable layers that extract the region of interest (ROI), which is the cancerous lesion. The method was evaluated on 3000 images–watershed masks, an augmented dataset. The model was trained on 2400 training set images and tested on 600 testing set images. This proposed method produced significant results of $98.53\%$ validation accuracy, $96.98\%$ validation dice coefficient, and $97.84\%$ validation intersection over unit (IoU) metric scores. Full article

Cell nucleus segmentation in microscopy images is an initial step in the quantitative analysis of imaging data, which is crucial for diverse biological and biomedical applications. While traditional machine learning methodologies have demonstrated limitations, recent advances in U-Net models have yielded promising improvements. However, it is noteworthy that these models perform well on balanced datasets, where the ratio of background to foreground pixels is equal. Within the realm of microscopy image segmentation, state-of-the-art models often encounter challenges in accurately predicting small foreground entities such as nuclei. Moreover, the majority of these models exhibit large parameter sizes, predisposing them to overfitting issues. To overcome these challenges, this study introduces a novel architecture, called mA-UNet, designed to excel in predicting small foreground elements. Additionally, a data preprocessing strategy inspired by road segmentation approaches is employed to address dataset imbalance issues. The experimental results show that the MIoU score attained by the mA-UNet model stands at $95.50\%$, surpassing the nearest competitor, UNet++, on the 2018 Data Science Bowl dataset. Ultimately, our proposed methodology surpasses all other state-of-the-art models in terms of both quantitative and qualitative evaluations. The mA-UNet model is also implemented in VHDL on the Zynq UltraScale+ FPGA, demonstrating its ability to perform complex computations with minimal hardware resources, as well as its efficiency and scalability on advanced FPGA platforms. Full article

Individual animal identification is a cornerstone of animal welfare practices and is of crucial importance for food safety and the protection of humans from zoonotic diseases. It is also a key prerequisite for enabling automated processes in modern dairy farming. With newly emerging technologies, visual animal identification based on machine learning offers a more efficient and non-invasive method with high automation potential, accuracy, and practical applicability. However, a common challenge is the limited variability of training datasets, as images are typically captured in controlled environments with uniform backgrounds and fixed poses. This study investigates the impact of foreground segmentation and background removal on the performance of convolutional neural networks (CNNs) for cow identification. A dataset was created in which training images of dairy cows exhibited low variability in pose and background for each individual, whereas the test dataset introduced significant variation in both pose and environment. Both a fine-tuned CNN backbone and a model trained from scratch were evaluated using images with and without background information. The results demonstrate that although training on segmented foregrounds extracts intrinsic biometric features, background cues carry more information for individual recognition. Full article

Accurate segmentation of the pancreas in Computed Tomography (CT) scans is a challenging task, which may be crucial for the diagnosis and treatment planning of pancreatic cancer. The irregular shape of the pancreas, its low contrast relative to surrounding tissues, and its close proximity to other complex anatomical structures make it difficult to accurately delineate its contours. Furthermore, a significant class imbalance between foreground (pancreas) and background voxels in an abdominal CT series represents an additional challenge for deep-learning-based approaches. In this study, we developed a deep learning model for automated pancreas segmentation based on a 3D U-Net architecture enhanced with an attention mechanism to improve the model capability to focus on relevant anatomical features of the pancreas. The model was trained and evaluated on two widely used benchmark datasets for volumetric segmentation, the NIH Healthy Pancreas-dataset and the Medical Segmentation Decathlon (MSD) pancreas dataset. The proposed attention-guided 3D U-Net achieved a Dice score of 80.8 ± 2.1%, ASSD of 2.1 ± 0.3 mm, and HD₉₅ of 8.1 ± 1.6 mm on the NIH dataset, and the values of 78.1 ± 1.1%, 3.3 ± 0.3 mm, and 12.3 ± 1.5 mm for the same metrics on the MSD dataset, demonstrating the value of attention mechanisms in improving segmentation performance in complex and low-contrast anatomical regions. Full article

An enhanced method called FreeViBe+ for moving target segmentation is proposed in this paper, addressing limitations in the ViBe algorithm such as ghosting, shadows, and holes. To eliminate ghosts, multi-frame background modeling is introduced. Shadows are detected and removed based on their characteristics in the HSV color space, while holes are filled by merging GrabCut segmentation results with the ViBe extraction output. Furthermore, the Structure-measure is tuned to optimize image fusion, enabling improved foreground–background separation. Comprehensive experiments on the UCF101 and Weizmann datasets demonstrate the effectiveness of FreeViBe+ in comparison with Finite Difference, Gaussian Mixture Model, and ViBe methods. Ablation studies confirm the individual contributions of multi-frame modeling, shadow removal, and GrabCut refinement, while sensitivity analysis verifies the robustness of key parameters. Quantitative evaluations show that FreeViBe+ achieves superior performance in precision, recall, and F-measure compared with existing approaches. Full article

With the rapid development of smart grids, deep power vision technologies are playing a vital role in monitoring the condition of transmission lines. In particular, for high-resolution and large-format transmission line images acquired during routine inspections, accurate extraction of transmission wires is crucial for efficient and accurate subsequent defect detection. In this paper, we propose a training-free (i.e., requiring no task-specific training or annotated datasets for wire extraction) wire extraction method specifically designed for large-scale transmission line images with complex backgrounds. The core idea is to leverage depth estimation maps to enhance the separation between foreground wires and complex backgrounds. This improved separability enables robust identification of slender wire structures in visually cluttered scenes. Building on this, a line segment structure-based method is developed, which identifies wire regions by detecting horizontally oriented linear features while effectively suppressing background interference. Unlike deep learning-based methods, the proposed method is training-free and dataset-independent. Experimental results show that our method effectively addresses background complexity and computational overhead in large-scale transmission line image processing. Full article

The high operational velocities of high-speed trains present constraints on their onboard track intrusion detection systems for real-time capture and analysis, encompassing limited computational resources and motion image blurring. This emphasizes the critical necessity of track perimeter intrusion monitoring systems. Consequently, an intelligent monitoring system employing trackside cameras is constructed, integrating weakly supervised video anomaly detection and unsupervised foreground segmentation, which offers a solution for monitoring foreign objects on high-speed train tracks. To address the challenges of complex dataset annotation and unidentified target detection, weakly supervised learning detection is proposed to track foreign object intrusions based on video. The pretraining of Xception3D and the integration of multiple attention mechanisms have markedly enhanced the feature extraction capabilities. The Top-K sample selection alongside the amplitude score/feature loss function effectively discriminates abnormal from normal samples, incorporating time-smoothing constraints to ensure detection consistency across consecutive frames. Once abnormal video frames are identified, a multiscale variational autoencoder is proposed for the positioning of foreign objects. A downsampling/upsampling module is optimized to increase feature extraction efficiency. The pixel-level background weight distribution loss function is engineered to jointly balance background authenticity and noise resistance. Ultimately, the experimental results indicate that the video anomaly detection model achieved an AUC of 0.99 on the track anomaly detection dataset and processes 2 s video segments in 0.41 s. The proposed foreground segmentation algorithm achieved an F1 score of 0.9030 in the track anomaly dataset and 0.8375 on CDnet2014, with 91 Frames per Second, confirming its efficacy. Full article