Prabhu Prasad Dev
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Prabhu Prasad Dev has a total of 11 co-authors. Here are the top co-authors by number of shared publications:
- Pranesh Das
Machine Learning Lab, Department of Computer Science and Engineering, National Institute of Technology Calicut, Kozhikode, India
2 shared publications Debi Prasad Senapati
1 shared publicationDipak Kumar Sahoo
Bhubaneswar Engineering College,Dept. of Computer Science and Engineering
1 shared publication- Hrudaya Kumar Tripathy
Kalinga Institute of Industrial Technology
1 shared publication Jyoti Prakash Sahoo
Siksha O Anusandhan (Deemed to be University),ITER,Dept of CSIT
1 shared publication- Lalit Mohan Satapathy
SOA Deemed to be University
1 shared publication
Latest Publications
- 2 January 2026
- Published by Elsevier in Neurocomputing
- Vol. 670
Cross-modal uncertainty modeling framework for unseen video anomaly detection
- 2 January 2026
- Published by Elsevier in Neurocomputing
- Vol. 670
Cross-modal uncertainty modeling framework for unseen video anomaly detection
- 2 July 2025
- Published by Elsevier in Journal of Visual Communication and Image Representation
- Vol. 111
CA-VAD: Caption Aware Video Anomaly Detection in surveillance videos
- 2 July 2025
- Published by Elsevier in Journal of Visual Communication and Image Representation
- Vol. 111
CA-VAD: Caption Aware Video Anomaly Detection in surveillance videos
- 21 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Video Technology
- Vol. PP (10518215) , 1
ReFLIP-VAD: Towards Weakly Supervised Video Anomaly Detection via Vision-Language Model
The vision-language model has recently achieved notable success in image-related tasks, showcasing its ability to learn deep and meaningful visual representations. Applying this robust model to video analysis for detecting anomalies poses a significant challenge. This paper introduces Reparameterized Fine-grained Language Image Pretraining-Video Anomaly Detection (ReFLIP-VAD), a novel approach designed to leverage vision-language capabilities for video anomaly detection. ReFLIP-VAD employs a prompt encoder to generate reparameterized learnable prompt templates, enhancing interpretability and understanding of anomaly-specific semantics. The framework adopts a dual-block architecture: a classification block that uses visual features for binary classification and a video-text alignment block that integrates textual and visual features for precise language-vision alignment. This proposed approach is further strengthened by the Glimpse-Emphasize network that effectively captures both global and local temporal dependencies across time and the MIL-Align mechanism that selects the most representative video frames for each label, representing the entire video. ReFLIP-VAD has demonstrated superior performance on two large-scale benchmark datasets achieving an Average Precision (AP) of 86.29% on XD-Violence and an Area Under the Curve (AUC) of 89.14% on UCF-Crime, significantly surpassing existing state-of-the-art methods. Our source code is available at https://github.com/prasaddev97/ReFLIP-VAD.
- 21 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Video Technology
- Vol. PP (10518215) , 1
ReFLIP-VAD: Towards Weakly Supervised Video Anomaly Detection via Vision-Language Model
The vision-language model has recently achieved notable success in image-related tasks, showcasing its ability to learn deep and meaningful visual representations. Applying this robust model to video analysis for detecting anomalies poses a significant challenge. This paper introduces Reparameterized Fine-grained Language Image Pretraining-Video Anomaly Detection (ReFLIP-VAD), a novel approach designed to leverage vision-language capabilities for video anomaly detection. ReFLIP-VAD employs a prompt encoder to generate reparameterized learnable prompt templates, enhancing interpretability and understanding of anomaly-specific semantics. The framework adopts a dual-block architecture: a classification block that uses visual features for binary classification and a video-text alignment block that integrates textual and visual features for precise language-vision alignment. This proposed approach is further strengthened by the Glimpse-Emphasize network that effectively captures both global and local temporal dependencies across time and the MIL-Align mechanism that selects the most representative video frames for each label, representing the entire video. ReFLIP-VAD has demonstrated superior performance on two large-scale benchmark datasets achieving an Average Precision (AP) of 86.29% on XD-Violence and an Area Under the Curve (AUC) of 89.14% on UCF-Crime, significantly surpassing existing state-of-the-art methods. Our source code is available at https://github.com/prasaddev97/ReFLIP-VAD.
- 8 October 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-6
An Enhanced Dynamic Image Watermarking Framework based on Phase Congruency and Saliency
Perceptual watermarking approaches achieve better image fidelity with the advantages of the Human Visual System (HVS). Saliency uses the advantages of this feature by identifying different visually perceptual regions of an image. Additionally, Phase congruency (PC) exploits those features of digital images by detecting the corner points. In the proposed method, the information hiding Capacity is generated from both bit depth maps created by comparing phase congruency results and saliency outcomes of an image that ascertains the low perceptually visible pixels (utilizing the Least Significant Bit (LSB) technique). The key idea is to ensure an additional layer of security using Symmetric Key Cryptography (SKC) and to hide information in a less observant area that is not easily perceptible inside the cover image. The experimental result analysis ensures better imperceptibility and robustness as it resists different attacks along with superior data hiding capacity.
- 8 October 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-6
An Enhanced Dynamic Image Watermarking Framework based on Phase Congruency and Saliency
Perceptual watermarking approaches achieve better image fidelity with the advantages of the Human Visual System (HVS). Saliency uses the advantages of this feature by identifying different visually perceptual regions of an image. Additionally, Phase congruency (PC) exploits those features of digital images by detecting the corner points. In the proposed method, the information hiding Capacity is generated from both bit depth maps created by comparing phase congruency results and saliency outcomes of an image that ascertains the low perceptually visible pixels (utilizing the Least Significant Bit (LSB) technique). The key idea is to ensure an additional layer of security using Symmetric Key Cryptography (SKC) and to hide information in a less observant area that is not easily perceptible inside the cover image. The experimental result analysis ensures better imperceptibility and robustness as it resists different attacks along with superior data hiding capacity.