Medical image segmentation plays a pivotal role in reducing radiologists’ workload by enabling accurate severity assessment and treatment planning. With coronary heart disease becoming an increasing concern globally, there is a growing need for efficient and reliable segmentation methods. In this context, we present TRACS, a self-supervised Triple Attention Network designed to enhance multi-scale feature representation for coronary vessel segmentation. The encoder generates dynamic channel-wise attention weights, effectively capturing fine structural details across multiple scales. This design is particularly well-suited for segmenting thin, elongated structures, such as vessels in angiograms, which are often challenging to delineate due to low contrast and occluded vessels. Unlike earlier self-supervised methods that rely on diffusion processes or multi-generator adversarial learning, which often introduce training complexities and optimization instabilities. TRACS adopts a streamlined architecture that not only ensures stable, consistent performance but also operates with a significantly reduced parameter count, making it a lightweight, efficient self-supervised architecture. The training process combines a fusion of losses to optimize both for regional accuracy and boundary precision. We evaluate TRACS on diverse unseen coronary angiograms (134XCA and 30XCA) and retinal vessel images (DRIVE and STARE). TRACS achieves significant inference gains in segmentation performance and inference speed, operating up to 3x faster than the nearest baseline method and with a significantly lower memory footprint, making it highly efficient for real-time applications and deployment on resource-constrained devices. We supplement the results with a thorough statistical analysis and explainability results from various Gradient and perturbation-based techniques. Results indicate that TRACS outperforms several recent self-supervised methods and matches the performance of supervised baselines in segmentation accuracy and robustness.TRACS offers a practical, domain-independent solution for vessel-structure segmentation in medical imaging. The code is publicly available on github .

Adversarial attacks pose a critical threat to the reliability of medical image classification in Industry 4.0, where smart, interconnected devices drive high-stakes diagnostic decisions. This article introduces Stealth, a simple yet highly optimized black-box adversarial attack specifically designed for medical imaging applications. Stealth is the first framework to utilize low-rank perturbation for adversarial generation. An optimization framework iteratively refines the complete singular value spectrum, reconstructs the perturbed image, and queries the target classifier to verify attack success—without relying on surrogate models or prediction probabilities. Improved imperceptibility is achieved by preserving visual fidelity throughout the perturbation process, ensuring the adversarial outputs remain indistinguishable from clean images. Operating as a true black-box method, Stealth requires only hard-label outputs, making it ideal for real-world deployments in Industry 4.0-enabled medical environments. Unlike existing approaches, it is entirely model-agnostic and image-modality independent, enabling broad applicability across healthcare systems. Extensive experiments across publicly available datasets—Chest X-Ray, Colonoscopy, MRI, and Mammogram—demonstrate that Stealth consistently achieves high attack success rates while surpassing state-of-the-art white-box attacks in perceptual quality, with explainability analysis further confirming its impact on model decision-making. These results expose critical vulnerabilities in current healthcare artificial intelligence (AI) systems and highlight the need for more robust adversarial defenses.

Despite their exceptional performance in the Internet of Medical Things (IoMT), deep learning models are susceptible to adversarial attacks. Existing defense approaches often suffer from limitations, including required model alterations, attack-type awareness, and high-computational complexity. This article introduces MedDefend: a lightweight, three-way detection technique, combining adaptive noise injection with a tailored robust principal component analysis (t-RPCA)-based noise mitigation. Initially, the method employs strategic image- dependent Gaussian noise injection, guided by class activation maps, to mask adversarial perturbations. Subsequently, t-RPCA is employed to eliminate the introduced noise along with the adversarial perturbations. Finally, the model evaluates classification consistency between original and denoised samples to detect potential adversarial examples. MedDefend effectively detects adversarial attacks across various medical imaging modalities, with a lightweight, training-free, and model-agnostic design suitable for IoMT integration. To encourage community engagement and reimplementation, our code is available athttps://github.com/nirmalpadichira/MedDefend/tree/main.

Class Activation Maps (CAMs) are one of the important methods for visualizing regions used by deep learning models. Yet their robustness to different noise remains underexplored. In this work, we evaluate and report the resilience of various CAM methods for different noise perturbations across multiple architectures and datasets. By analyzing the influence of different noise types on CAM explanations, we assess the susceptibility to noise and the extent to which dataset characteristics may impact explanation stability. The findings highlight considerable variability in noise sensitivity for various CAMs. We propose a robustness metric for CAMs that captures two key properties: consistency and responsiveness. Consistency reflects the ability of CAMs to remain stable under input perturbations that do not alter the predicted class, while responsiveness measures the sensitivity of CAMs to changes in the prediction caused by such perturbations. The metric is evaluated empirically across models, different perturbations, and datasets along with complementary statistical tests to exemplify the applicability of our proposed approach.

In this article, we propose a robust algorithm for the fast recovery of human actions from corrupted 3-D motion capture (mocap) sequences. The proposed algorithm can deal with misrepresentations and incomplete representations in mocap data simultaneously. Fast convergence of the proposed algorithm is ensured by minimizing the overhead associated with time and resource utilization. To this end, we have used an approximate singular value decomposition (SVD) based on QR decomposition and$\ell _{2,1}$norm minimization as a replacement for the conventional nuclear norm-based SVD. In addition, the proposed method is braced by incorporating the spatio-temporal properties of human action in the optimization problem. For this, we have introduced pair-wise hierarchical constraint and the trajectory movement constraint in the problem formulation. Finally, the proposed method is void of the requirement of a sizeable database for training the model. The algorithm can easily be adapted to work on any form of corrupted mocap sequences. The proposed algorithm is faster by 30% on average compared with the counterparts employing similar kinds of constraints with improved performance in recovery.

Segmenting meaningful visual structures from an image is a fundamental and most-addressed problem in image analysis algorithms. However, among factors such as diverse visual patterns, noise, complex backgrounds, and similar textures present in foreground and background, image segmentation still stands as a challenging research problem. In this article, the proposed method employs an unsupervised method that addresses image segmentation as subspace clustering of image feature vectors. Initially, an image is partitioned into a set of homogeneous regions called superpixels, from which Local Spectral Histogram features are computed. Subsequently, a feature data matrix is created whereupon subspace clustering methodology is applied. A single-stage optimization model is formulated with enhanced segmentation capabilities by the combined action of l ½ and l 2 norm minimization. Robustness of l ½ regularization toward both the noise and overestimation of sparsity provides simultaneous noise robustness and better subspace selection, respectively. While l 2 norm facilitates grouping effect. Hence, the designed optimization model ensures an improved sparse solution and a sparse representation matrix with an accurate block diagonal structure, which thereby favours getting properly segmented images. Then, experimental results of the proposed method are compared with the state-of-art algorithms. Results demonstrate the improved performance of our method over the state-of-art algorithms.

Nowadays, intelligent transport systems have a major role in providing a safe and secure traffic society for passengers, pedestrians, and vehicles. However, some bad weather conditions such as haze or fog may affect the visual clarity of video footage captured by the camera. This will cause a malfunction in further video processing algorithms performed by such automated systems. This article proposes an efficient technique for estimating the atmospheric light and the transmission map in the haze model entirely in tensor domain for video dehazing. In this work, the atmospheric light is appraised using the Mie scattering principle of visible light and the temporal coherency among the frames is achieved by means of tensor algebra. Furthermore, the transmission map is computed using Low Rank Tensor Approximation (LRTA) based on Weighted Tensor Nuclear Norm (WTNN) minimization and Tensor Total Variation (TTV) regularization. WTNN minimization is used to smooth the coarse transmission map, and TTV regularization is employed to maintain spatio-temporal continuity by preserving the details of salient structures and edges. The novelty of the proposed model is confined in the efficient formulation of a unified optimization model for the estimation of transmission map and atmospheric light in the tensor domain with fine-tuned regularization terms, which is not reported till now in the direction of video dehazing. Extensive experiments show that the proposed method outperforms state-of-the-art methods in video dehazing.

The rising demand for surveillance systems naturally necessitates more efficient and noise robust moving object detection (MOD) systems from the captured video streams. Inspired by the challenges in MOD which are yet to be addressed properly, this paper proposes a new MOD scheme using l1/2regularization in the tensor framework. It takes advantage of the special features of tensor singular value decomposition (t-SVD) along with regularizations using l1/2-norm with half thresholding operation and tensor total variation (TTV) to develop a noise robust MOD system with improved detection accuracy. While t -SVD exploits the spatio-temporal correlation of the video background, l1/2regularization provides noise robustness besides removing the sparser but discontinuous dynamic elements in the spatio-temporal direction. Moreover, TTV enhances the spatio-temporal continuity and fills up the gaps due to the lingering objects and thereby extracting the foreground precisely. The proposed three-way optimization method is designed to address both static and dynamic background cases of MOD separately with the intention to reduce the misclassifications due to moving/cluttered background. The brilliance of this method is confirmed by the impressive visual quality of the background/foreground separation, noise robustness, reduced computational complexity, and rapid response. The quantitative evaluation discloses the predominance of the proposed method with respect to the state-of-the-art techniques.

Outdoor monitoring systems are known to exhibit better performance under normal weather conditions, while it lacks effectiveness under inclement conditions. Often video footage captured by the camera under rainy conditions comprises several visual distortions. It eventually leads to flaws when handled with succeeding computer vision algorithms, namely the object identification and tracking. Additionally, eliminating such unpleasant rainy effects is essential prior to the processing of video footage by suitable algorithms. The present work attempts to formulate a new low‐rank tensor recovery based deraining algorithm that enables to remove the rain streaks from video footage. The proposed method detects the rain streaks by adopting optical flow estimation along with the brightness features inherent with the rain streaks. A unified framework comprised of tensor singular value decomposition (t‐SVD) based weighted nuclear norm minimisation and tensor total variation (TTV) regularisation effectively removes rain streaks and recovers the original rain‐free data from the available rainy data. The use of t‐SVD enforces the concept of low rankness and also exploits the temporal redundancy among the video frames. Furthermore, TTV regularisation facilitates to promote the temporal continuity for discriminating most of the natural image contents from sparse rain streaks by preserving piece‐wise smoothness of video frames. Comprehensive experimental findings based on real and synthetic data with dynamic background show that the rain streaks are more efficaciously eliminated by adopting the proposed method without much loss in the information.

Image super resolution refers to a class of signal processing algorithms to post‐process a captured image to obtain its high resolution version. Multi‐frame super resolution synthesises high resolution image from multiple low resolution observations. Performance of super resolution algorithms are adversely affected by the noise present in the input images. To develop a noise robust multi‐frame image super resolution, an objective function is formulated which contains a weighted data fidelity term and a regularisation term consisting of a bilateral total variation (BTV) term and structure tensor total variation (STV) term. Both BTV and STV are weighted appropriately in a per pixel basis in such a way that the BTV contributes more in smooth regions and STV contributes more on the edges. These terms ensure the continuity of edges and the smoothness of flat regions. An adaptive weighting scheme with the data fidelity term helps to select the reliable pixel alone in the reconstruction process. The proposed method is experimentally evaluated for its performance in real data and different types of noises.

In this paper, entropy of singular values is used to promote the low rank property which accounts the inherent spatial redundancy and spectral correlation. Inspired from the connection between randomness and compactness of signals, we minimise ENtropy for low rank Tensor Completion (ENTC) using tensor-Singular Value Decomposition (t-SVD) framework. This approach assures better performance in excessive loss and complex low rank structure scenarios. As opposed to state-of- the-art-methods, the proposed approach outperforms with a fewer number of reliable samples and extends the region of recovery. The significant performance improvement is observed in terms of Inverse Relative Squared Error (iRSE), Average Structural SIMilarity (ASSIM), and Average Feature SIMilarity (AFSIM) measures.

The spatially distributed digital cameras in Wireless Multimedia Sensor Networks (WMSN) are provided with miniature batteries resulting in power constraints. These cameras acquire compressive measurements of video at a rate significantly below the Nyquist rate and transmit them wirelessly in the network. Thus, the encoding side (transmitter) is made less complex at the expense of increased complexity at the more resourceful decoder (receiver). Well grounded on this relevant practical scenario, a unified system is proposed that integrates detection of moving objects into the Compressive Sensing (CS) recovery framework and thereby realizing simultaneous data recovery and object detection in a single optimization problem which guarantees fast response. In this work, a new tensor RPCA approach is proposed to accomplish this requirement. For background separation, low rank approximation is done on the highly correlated background components. A Laplace function based surrogate for tensor tubal rank is formulated to provide adaptive thresholding for the singular value tubes of the background tensor. Moreover, the spatio-temporal continuity of the foreground is explored using 3D-Piecewise Smoothness Constraints combinations based Anisotropic Total Variation (3D-PSCATV) regularization. Additionally, $l_{1}$ regularization has been adopted to describe the sparsity of moving objects. The proposed model is solved using Alternative Direction Method of Multipliers (ADMM) scheme. The quantitative and qualitative results validate the superior performance of the proposed method against the compared approaches.

Incomplete observations in the data are always troublesome to data clustering algorithms. In fact, most of the well-received techniques are not designed to encounter such imperative scenarios. Hence, clustering of images under incomplete samples is an inquisitive yet unaddressed area of research. Therefore, the aim of this article is to design a single-stage optimization procedure for clustering as well as simultaneous reconstruction of images without breaking the intrinsic spatial structure. The method employs the self-expressiveness property of submodules, and images are stacked as the lateral slices of a three-dimensional tensor. The proposed optimization method is designed to extract a sparse t -linear combination tensor with low multirank constraint, consisting of a unique set of linear coefficients in the form of mode-3 fibers and the spectral clustering is performed on these fibers. Simultaneously, the recovery of lost samples is accomplished by twisting the entire lateral slices of the data tensor and applying a low-rank approximation on each slice. The prominence of the proposed method lies in the simultaneous execution of data clustering and reconstruction of incomplete observations in a single step. Experimental results reveal the excellence of the proposed method over state-of-the-art clustering algorithms in the context of incomplete imaging data.

Since automated cleaning systems are less common in extreme surveillance environments, the accumulations of combustion fuels, dust, dirt, spider webs, etc., affect the visibility and clarity of the captured video data to different degrees resulting in incomplete (missing) video sequences. Grounded on this significant practical scenario, this letter proposes a scheme to concurrently complete the missing entries and detect moving objects with efficient background separation by formulating a single convex optimization problem developed and implemented in the tensor framework. The work is implemented as a unified scheme for concurrent video completion and moving object detection using twist spatio-temporal total variation to enhance the detection performance of the foreground while fitting the tensor nuclear norm minimization for efficient background separation with half thresholding applied on the tensor singular values. Moreover, the sparse variations that are part of the dynamic background are addressed using $l_{1/2}$ regularization. The formulated minimization problem is solved using the augmented Lagrangian method with an alternating direction strategy. The work also has computational benefits and the excellence of this method is revealed in the quantified performance evaluation against the compared approaches.

This paper focuses on an unsupervised anomaly detection approach from a subsampled hyperspectral image (HSI) data. Unlike the state-of-the-art methods, the proposed method does not require the construction of a large matrix by the vectorization of spectral bands or unfolding of entire HSI data into a huge matrix. Image reconstruction and anomaly detection are performed at the same time on the subsampled HSI data to reduce the storage/transmission requirements and processing time. In the proposed framework, the HSI data are decomposed into background and anomaly by l(1/2)regularized joint sparse and low-rank decomposition. Since the spectral bands of HSI data are highly correlated, it can be effectively utilized for background modeling. Inspired by this fact, the low-dimensional structure of the background is characterized by modeling it as a combination of common low-rank and spectral-specific low-rank components. To further improve the separation between background and anomaly, the recently proposed l(1/2)regularization for the sparse component is employed. The experimental results reveal that the proposed method outperforms the existing methods on both reconstruction quality and detection performance.