Institution profile

Northumbria University

Academic institutioneurope · gb
Official website
Research library54linked papers
Opportunities0open roles
Selected work

Representative Papers

Video Forgery Detection for Surveillance Cameras: A Review

May 04, 2025

To address the growing threat of video tampering in surveillance footage—which undermines its admissibility as judicial evidence—this paper presents a systematic survey of video forgery detection techniques tailored to security monitoring scenarios. We propose the first robustness evaluation framework specifically designed for real-world surveillance conditions, characterized by low resolution, high compression, and dynamic illumination variations. The framework integrates compression artifact analysis, temporal consistency verification, and hybrid feature extraction combining deep learning models (CNNs and LSTMs) with handcrafted features. For the first time, we conduct a comprehensive comparative analysis of three mainstream approaches—compression-based feature analysis, frame duplication detection, and machine learning–based methods—elucidating their respective applicability boundaries and performance limitations under practical surveillance constraints. Our empirical study identifies characteristic failure modes of existing detectors across typical surveillance conditions, thereby providing evidence-based guidance for forensic system design, algorithm optimization, and standardization efforts in digital video authentication.

1 citationsRead paper

Negotiating Risk Boundaries in AI for Policing Through Mixed-Stakeholder Deliberation

Aug 05, 2026

This study addresses fairness risks in current AI applications in policing, which often arise from the exclusion of communities disproportionately affected by racial bias. By convening 30 community members, police officers, and scholars in a mixed-stakeholder deliberative workshop, the research conducts a systematic risk–benefit analysis of 13 AI use cases in law enforcement, uniquely integrating a racial equity lens into the evaluation framework from the outset. Combining qualitative dialogue with structured risk assessment, the findings reveal that inclusive deliberation effectively steers participants toward prioritizing social efficacy and equitable impact over mere technical feasibility. While most use cases were broadly accepted, participants explicitly rejected three high-risk applications, including recidivism risk assessment. The deliberative process exhibited “curb-cut effect”–style integrative reasoning, fostering a more equitable and consensus-driven pathway for AI governance.

0 citationsRead paper

Adaptive Contrast Enhancement and Optimised Feature Matching for RootSIFT-Based Palm-Vein Recognition

Jul 17, 2026

This study addresses the degraded recognition performance of low-contrast palm vein images caused by near-infrared scattering and sensor limitations. The authors propose an enhanced method, ILACS-BGOT, which improves local contrast while effectively suppressing block artifacts. They further develop a recognition pipeline integrating RootSIFT features, KNN combined with Random Sample Consensus (RANSAC)-based transformation (RT) matching, and Mean and Median Distance (MMD) filtering. Notably, the original ILACS-LGOT is innovatively refined into ILACS-BGOT to better preserve fine details. The work also provides a systematic analysis of how MMD and RT parameters influence cross-dataset generalization. Evaluated on the CASIA, PolyU, and PUT datasets, the proposed approach significantly outperforms existing methods, achieving substantially lower equal error rates (EER) and higher accuracy, with performance consistently improving as template size increases.

0 citationsRead paper
Recent publications

Latest Papers

Negotiating Risk Boundaries in AI for Policing Through Mixed-Stakeholder Deliberation

Aug 05, 2026

This study addresses fairness risks in current AI applications in policing, which often arise from the exclusion of communities disproportionately affected by racial bias. By convening 30 community members, police officers, and scholars in a mixed-stakeholder deliberative workshop, the research conducts a systematic risk–benefit analysis of 13 AI use cases in law enforcement, uniquely integrating a racial equity lens into the evaluation framework from the outset. Combining qualitative dialogue with structured risk assessment, the findings reveal that inclusive deliberation effectively steers participants toward prioritizing social efficacy and equitable impact over mere technical feasibility. While most use cases were broadly accepted, participants explicitly rejected three high-risk applications, including recidivism risk assessment. The deliberative process exhibited “curb-cut effect”–style integrative reasoning, fostering a more equitable and consensus-driven pathway for AI governance.

0 citationsRead paper

Adaptive Contrast Enhancement and Optimised Feature Matching for RootSIFT-Based Palm-Vein Recognition

Jul 17, 2026

This study addresses the degraded recognition performance of low-contrast palm vein images caused by near-infrared scattering and sensor limitations. The authors propose an enhanced method, ILACS-BGOT, which improves local contrast while effectively suppressing block artifacts. They further develop a recognition pipeline integrating RootSIFT features, KNN combined with Random Sample Consensus (RANSAC)-based transformation (RT) matching, and Mean and Median Distance (MMD) filtering. Notably, the original ILACS-LGOT is innovatively refined into ILACS-BGOT to better preserve fine details. The work also provides a systematic analysis of how MMD and RT parameters influence cross-dataset generalization. Evaluated on the CASIA, PolyU, and PUT datasets, the proposed approach significantly outperforms existing methods, achieving substantially lower equal error rates (EER) and higher accuracy, with performance consistently improving as template size increases.

0 citationsRead paper

comprisk: A scikit-learn-compatible Python toolkit for competing-risks survival analysis

Jul 10, 2026

Medical time-to-event data are often subject to competing risks, and conventional survival analysis methods introduce bias by treating competing events as censored observations. This work proposes the first Python-based competing risks analysis toolkit compatible with scikit-learn, offering a unified implementation of Fine–Gray regression, cause-specific Cox models, Aalen–Johansen cumulative incidence estimation, Gray’s test, and competing risks random survival forests. The package supports specialized evaluation metrics, including time-dependent AUC, Brier score, cause-specific concordance index, and calibration curves. Leveraging Numba acceleration, histogram-based splitting, and inverse probability of censoring weighting, the toolkit achieves 10–22 times faster training on million-scale electronic health records compared to R’s randomForestSRC, while maintaining numerical accuracy as verified through rigorous validation.

0 citationsRead paper