Sourav Saha
Sourav Saha has a total of 52 co-authors. Here are the top co-authors by number of shared publications:
- Biswajit Halder
Hooghly Engineering and Technology College Hooghly India
4 shared publications Priya Ranjan Sinha Mahapatra
Kalyani University
4 shared publications- Priya Ranjan Sinha Mahapatra
University of Kalyani
4 shared publications - Debjani Chakraborty
Narula Institute of Technology Kolkata India
3 shared publications - Priya Ranjan Sinha Mahapatra
University of Kalyani
3 shared publications - Shambo Chatterjee
University of Kalyani
3 shared publications
Latest Publications
- 1 May 2026
- Published by Springer Nature
- p. 239-254
Duplication Fraudulent Detection of Tampered Image Using Optimized CNN
- 1 May 2026
- Published by Springer Nature
- p. 239-254
Duplication Fraudulent Detection of Tampered Image Using Optimized CNN
- 2 January 2026
- Published by Springer Nature
- p. 139-154
Prediction of Soil Shear Strength Parameters at Different Depths of Missing Borehole Profile Using Machine Learning Model
- 2 January 2026
- Published by Springer Nature
- p. 139-154
Prediction of Soil Shear Strength Parameters at Different Depths of Missing Borehole Profile Using Machine Learning Model
- 2 January 2026
- Published by Springer Nature
- p. 101-114
Estimation of Missing Shear Strength Parameters of a Borehole Profile Using Machine Learning Model
- 2 January 2026
- Published by Springer Nature
- p. 101-114
Estimation of Missing Shear Strength Parameters of a Borehole Profile Using Machine Learning Model
- 28 October 2025
- Published by Wiley in Journal of Forensic Sciences
- Vol. 71 (1) , 388-404
Learning to localize image forgery using boundary‐preserving mask R‐ CNN
Digital image manipulation is a growing concern in multimedia security. Many existing forgery detection methods struggle with accurately localizing manipulated regions—especially near boundaries—and often fail to generalize across different manipulation types. To address these challenges, we propose a novel Boundary‐Preserving Mask R‐CNN framework that enhances detection precision by incorporating boundary‐aware features. The model integrates channel attention mechanisms to better capture detailed spatial information and leverages frequency domain features to improve robustness. We evaluated the framework on six diverse benchmark datasets—CASIA V2, Columbia, Carvalho, CoMoFoD, MICC‐F220, and CG‐1050—covering splicing, copy‐move, and compositing manipulations. Extensive preprocessing ensured uniform input, and pixel‐level segmentation enabled accurate region detection. Our method demonstrated strong performance across multiple metrics, including accuracy, precision, recall, F1‐score, IoU, and AUC. These results highlight its potential as a reliable tool for digital forensics and investigative applications.
- 28 October 2025
- Published by Wiley in Journal of Forensic Sciences
- Vol. 71 (1) , 388-404
Learning to localize image forgery using boundary‐preserving mask R‐ CNN
Digital image manipulation is a growing concern in multimedia security. Many existing forgery detection methods struggle with accurately localizing manipulated regions—especially near boundaries—and often fail to generalize across different manipulation types. To address these challenges, we propose a novel Boundary‐Preserving Mask R‐CNN framework that enhances detection precision by incorporating boundary‐aware features. The model integrates channel attention mechanisms to better capture detailed spatial information and leverages frequency domain features to improve robustness. We evaluated the framework on six diverse benchmark datasets—CASIA V2, Columbia, Carvalho, CoMoFoD, MICC‐F220, and CG‐1050—covering splicing, copy‐move, and compositing manipulations. Extensive preprocessing ensured uniform input, and pixel‐level segmentation enabled accurate region detection. Our method demonstrated strong performance across multiple metrics, including accuracy, precision, recall, F1‐score, IoU, and AUC. These results highlight its potential as a reliable tool for digital forensics and investigative applications.
- 15 July 2025
- Published by Springer Nature
- p. 353-366
A Two-Stage Autoencoder-Based Satellite Image Analysis Framework for Green Land Cover Detection
- 15 July 2025
- Published by Springer Nature
- p. 353-366
A Two-Stage Autoencoder-Based Satellite Image Analysis Framework for Green Land Cover Detection
- 12 February 2025
- Published by Springer Nature
- p. 94-109
A Computer Vision Model Utilizing Autoencoders for Surface Defect Recognition
- 12 February 2025
- Published by Springer Nature
- p. 94-109
A Computer Vision Model Utilizing Autoencoders for Surface Defect Recognition
- Open Access
- 8 February 2025
- Published by MECS Publisher in International Journal of Image, Graphics and Signal Processing
- Vol. 17 (1) , 45-58
An Autoencoder-Based Deep Learning Model forFractographic Characterization of TungstenHeavy Alloys
Fracture surface analysis is crucial in investigating manufacturing failures and material characterization. Traditional manual inspection methods are slow and subjective, prompting the need for efficient automated tools using advanced computer vision techniques. Recent machine learning models for classifying surface fractures show potentials but struggle due to the lack of large, labeled datasets. This study explores the potential application of autoencoders, a self-supervised neural network, to identify unintended fracture surfaces from anomalous manufacturing of tungstenheavy alloys. The proposed autoencoder-based model achieves 97% accuracy in distinguishing undesirable fracture patterns by analyzing the reconstruction loss of the images, surpassing existing methods. This high accuracy highlights the autoencoder's ability to automatically extract and reduce dimensional features from fracture surfaces effectively. The experimental result obtained on tungsten-heavy alloys demonstrate the model's potential towards developing autoencoder-based automated tools for fractographic analyses across various materials and operational scenarios.
- 8 February 2025
- Published by MECS Publisher in International Journal of Image, Graphics and Signal Processing
- Vol. 17 (1) , 45-58
Open AccessAn Autoencoder-Based Deep Learning Model forFractographic Characterization of TungstenHeavy Alloys
Fracture surface analysis is crucial in investigating manufacturing failures and material characterization. Traditional manual inspection methods are slow and subjective, prompting the need for efficient automated tools using advanced computer vision techniques. Recent machine learning models for classifying surface fractures show potentials but struggle due to the lack of large, labeled datasets. This study explores the potential application of autoencoders, a self-supervised neural network, to identify unintended fracture surfaces from anomalous manufacturing of tungstenheavy alloys. The proposed autoencoder-based model achieves 97% accuracy in distinguishing undesirable fracture patterns by analyzing the reconstruction loss of the images, surpassing existing methods. This high accuracy highlights the autoencoder's ability to automatically extract and reduce dimensional features from fracture surfaces effectively. The experimental result obtained on tungsten-heavy alloys demonstrate the model's potential towards developing autoencoder-based automated tools for fractographic analyses across various materials and operational scenarios.
- Open Access
- 1 December 2024
- Published by Walter de Gruyter GmbH in Journal of Information Assurance and Security
- Vol. 19 (4) , 146-161
A Two-stage CNN Based Computer Vision Framework for Automated Validation of Indian Bank Cheques
Automated bank cheque processing is still considered a challenging task for computer vision researchers. This article proposes a two-stage deep learning-based end-to-end computer vision framework to validate an Indian bank cheque with respect to a few mistakes commonly found in manually entered handwritten fields. The proposed framework primarily works in two stages involving two separate Mask RCNN models to detect two common mistakes due to the absence of any key handwritten field or the presence of any overwritten/strike-through handwritten character in the bank cheque image. The first stage of the Mask RCNN model aims to segment all the key handwritten fields in a bank cheque image, leading to the detection of any missing handwritten field. The second stage, the Mask RCNN model, attempts to detect the presence of any overwritten/strike-through handwritten character in a bank cheque image that may lead to the invalidation of the cheque. Due to the unavailability of any standard dataset for validation purposes, a bank cheque image data repository has been prepared exclusively for developing the proposed framework. Extensive experimentation with the prepared dataset reveals that the proposed framework can outperform some of the popular frameworks by achieving a promising accuracy (98%) in terms of reporting validation errors owing to the aforementioned mistakes in the bank cheque.
- 1 December 2024
- Published by Walter de Gruyter GmbH in Journal of Information Assurance and Security
- Vol. 19 (4) , 146-161
Open AccessA Two-stage CNN Based Computer Vision Framework for Automated Validation of Indian Bank Cheques
Automated bank cheque processing is still considered a challenging task for computer vision researchers. This article proposes a two-stage deep learning-based end-to-end computer vision framework to validate an Indian bank cheque with respect to a few mistakes commonly found in manually entered handwritten fields. The proposed framework primarily works in two stages involving two separate Mask RCNN models to detect two common mistakes due to the absence of any key handwritten field or the presence of any overwritten/strike-through handwritten character in the bank cheque image. The first stage of the Mask RCNN model aims to segment all the key handwritten fields in a bank cheque image, leading to the detection of any missing handwritten field. The second stage, the Mask RCNN model, attempts to detect the presence of any overwritten/strike-through handwritten character in a bank cheque image that may lead to the invalidation of the cheque. Due to the unavailability of any standard dataset for validation purposes, a bank cheque image data repository has been prepared exclusively for developing the proposed framework. Extensive experimentation with the prepared dataset reveals that the proposed framework can outperform some of the popular frameworks by achieving a promising accuracy (98%) in terms of reporting validation errors owing to the aforementioned mistakes in the bank cheque.
- 28 July 2024
- Published by Springer Nature
- p. 15-30
A Two-Stage CNN Based Satellite Image Analysis Framework for Estimating Building-Count in Residential Built-Up Area
- 28 July 2024
- Published by Springer Nature
- p. 15-30
A Two-Stage CNN Based Satellite Image Analysis Framework for Estimating Building-Count in Residential Built-Up Area