Aditya Nigam
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
Research Subjects
Research Interests
Aditya Nigam has a total of 85 co-authors. Here are the top co-authors by number of shared publications:
- Arnav Bhavsar
Indian Institute of Technology Mandi
10 shared publications Daksh Thapar
Indian Institute of Technology
5 shared publications- Aman Kamboj
National Institute of Technology Jalandhar, Jalandhar, India
4 shared publications - Avantika Singh
IIIT Naya Raipur
4 shared publications - Rajneesh Rani
Department of Computer Science and Engineering, NIT, Jalandhar, India
4 shared publications - Ranjeet Ranjan Jha
MANAS Lab, School of Computing and Electrical Engineering (SCEE), Indian Institute of Technology (IIT) Mandi, India
4 shared publications
Latest Publications
- Open Access
- 20 September 2026
- Published by Elsevier in SSRN Electronic Journal
Information-Theoretic Biometric Analysis of Forehead Crease Patterns for Identity Verification
Conventional biometric traits such as the face, iris, and fingerprint are well established, whereas the biometric potential of subtle facial characteristics remains largely unexplored. Forehead creases remain visible under partial facial occlusion and exhibit structured patterns that may encode subject-specific information; however, their biometric distinctiveness remains unclear. To our knowledge, this is among the first systematic information-theoretic analyses of forehead creases for biometric verification. We propose DINOv2-FCR, a dual-stream foundation-model framework that combines RGB appearance with Laplacian-enhanced high-frequency information. Beyond verification, we quantify subject specific information using KL divergence and normalized information density. Extensive open-set experiments on unconstrained forehead data show that DINOv2-FCR achieves an EER of 8.32%, with TMRs of 59.76% at 0.1% FMR and 42.68% at 0.01% FMR, with differencesfrom CNN and transformer baselines largely within bootstrap uncertainty. The information theoretic analysis further shows measurable subject specific information, with DINOv2-FCR achieving 0.64 bits/dim while using a compact 256 dimensional embedding. These results establish forehead creases as a weak but detectable source of identity related information and demonstrate the potential of foundation model representations for forehead based biometric verification.
- 20 September 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessInformation-Theoretic Biometric Analysis of Forehead Crease Patterns for Identity Verification
Conventional biometric traits such as the face, iris, and fingerprint are well established, whereas the biometric potential of subtle facial characteristics remains largely unexplored. Forehead creases remain visible under partial facial occlusion and exhibit structured patterns that may encode subject-specific information; however, their biometric distinctiveness remains unclear. To our knowledge, this is among the first systematic information-theoretic analyses of forehead creases for biometric verification. We propose DINOv2-FCR, a dual-stream foundation-model framework that combines RGB appearance with Laplacian-enhanced high-frequency information. Beyond verification, we quantify subject specific information using KL divergence and normalized information density. Extensive open-set experiments on unconstrained forehead data show that DINOv2-FCR achieves an EER of 8.32%, with TMRs of 59.76% at 0.1% FMR and 42.68% at 0.01% FMR, with differencesfrom CNN and transformer baselines largely within bootstrap uncertainty. The information theoretic analysis further shows measurable subject specific information, with DINOv2-FCR achieving 0.64 bits/dim while using a compact 256 dimensional embedding. These results establish forehead creases as a weak but detectable source of identity related information and demonstrate the potential of foundation model representations for forehead based biometric verification.
- 1 September 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 26 (18) , 28168-28182
UniMoRe : Uni fied Mo tion Re presentation via Gait IMU Pretraining
- 1 September 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 26 (18) , 28168-28182
UniMoRe : Uni fied Mo tion Re presentation via Gait IMU Pretraining
- 3 August 2026
- Published by Springer Nature
- p. 553-568
OdorNet: An Approach for Smell Digitization and Classification
- 3 August 2026
- Published by Springer Nature
- p. 553-568
OdorNet: An Approach for Smell Digitization and Classification
- 3 February 2026
- Published by Springer Nature
- p. 343-357
Bh$\bar{a}$luk: Learning the Unknown Basis of Human Olfactory System Using Deep Learning- 3 February 2026
- Published by Springer Nature
- p. 343-357
Bh$\bar{a}$luk: Learning the Unknown Basis of Human Olfactory System Using Deep Learning- Open Access
- 17 December 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-7
On the Transferability of LaBraM: Evaluating Representations across Visual and Clinical Domains with Convolutional Adapters
Electroencephalography (EEG) is widely used in brain-computer interface (BCI) applications due to its non-invasive nature and high temporal resolution. However, the heterogeneity of EEG acquisition protocols, device configurations, and task-specific setups presents a major challenge for developing general-purpose EEG models. Recent foundational models, such as LaBraM, offer the potential to learn transferable EEG representations from large-scale data. In this work, we systematically evaluate the downstream utility of LaBraM across two distinct EEG domains: visual perception and clinical diagnosis. Due to variability in EEG configurations across datasets, directly applying LaBraM representations is challenging, necessitating specialized adapters. To adapt LaBraM for classification tasks, we design lightweight convolutional adapters that require minimal fine-tuning. Our framework is assessed on two public datasets: EEG-ImageNet, which involves stimulus-based visual decoding, and BrainLat, a resting-state dataset for neurodegenerative disease classification. The proposed approach achieves state-of-the-art performance, with 99.33 \(\%\) accuracy on EEG-ImageNet and 92.31 \(\%\) subject-level accuracy on BrainLat under a subject-independent setting. Additionally, we demonstrate generalization to unseen classes via zero-shot classification. These results highlight the practical transferability of LaBraM’s representations and provide insights for designing EEG systems that span both cognitive and clinical applications.- 17 December 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-7
Open AccessOn the Transferability of LaBraM: Evaluating Representations across Visual and Clinical Domains with Convolutional Adapters
Electroencephalography (EEG) is widely used in brain-computer interface (BCI) applications due to its non-invasive nature and high temporal resolution. However, the heterogeneity of EEG acquisition protocols, device configurations, and task-specific setups presents a major challenge for developing general-purpose EEG models. Recent foundational models, such as LaBraM, offer the potential to learn transferable EEG representations from large-scale data. In this work, we systematically evaluate the downstream utility of LaBraM across two distinct EEG domains: visual perception and clinical diagnosis. Due to variability in EEG configurations across datasets, directly applying LaBraM representations is challenging, necessitating specialized adapters. To adapt LaBraM for classification tasks, we design lightweight convolutional adapters that require minimal fine-tuning. Our framework is assessed on two public datasets: EEG-ImageNet, which involves stimulus-based visual decoding, and BrainLat, a resting-state dataset for neurodegenerative disease classification. The proposed approach achieves state-of-the-art performance, with 99.33 \(\%\) accuracy on EEG-ImageNet and 92.31 \(\%\) subject-level accuracy on BrainLat under a subject-independent setting. Additionally, we demonstrate generalization to unseen classes via zero-shot classification. These results highlight the practical transferability of LaBraM’s representations and provide insights for designing EEG systems that span both cognitive and clinical applications. - Open Access
- 17 December 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-8
EEG-Guided Image Synthesis via Hybrid Embedding and Diffusion Framework
- 17 December 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-8
Open AccessEEG-Guided Image Synthesis via Hybrid Embedding and Diffusion Framework
- Open Access
- 6 December 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-7
PENFORMER: Identifying Signature Truth with Log Normal Aided Multimodal Transformer
- 6 December 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-7
Open AccessPENFORMER: Identifying Signature Truth with Log Normal Aided Multimodal Transformer
- Open Access
- 10 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 101472-101490
FISNET: A Learnable Fusion-Based Iris Segmentation Network Improving Robustness Across NIR and VIS Modalities
Iris is one of the most distinctive biometric traits used for reliable identity verification in applications such as border control, secure access systems, and national ID programs. The primary challenge in iris recognition is the reliable segmentation of the iris region from the eye image. Segmenting irises captured under different imaging conditions is challenging for a single model due to variations in spectral features, texture, lighting conditions, and noise patterns between NIR and VIS images. To tackle this problem, we present Fused Iris Segmentation Network (FISNET) that combines segmentation maps from two models to achieve enhanced precision and accuracy. FISNET demonstrates robust generalizability across varying lighting, resolution, and sensor types, consistently outperforming individual models on all NIR and VIS datasets, including smartphone-captured images. The performance of FISNET was evaluated on CASIA-V4 subsets, Dark and Blue iris datasets, and the AFHIRIS dataset, achieving superior segmentation accuracy and recognition performance. The results demonstrate significant improvements over the IrisParseNet,PixlSegNet, and SAM model, achieving remarkable mIoU scores of 0.955, 0.930, 0.945, 0.955, 0.907, 0.815, 0.924, 0.913, 0.842, 0.852, 0.829, and 0.839 on the Lamp-V4, Interval-V4, Thousand-V4, Syn-V4, Twins-V4, UBIRIS-V2, BI-P1, BI-P2, DI-P1, DI-P2, DI-P3, and AFHIRIS-V1 datasets, respectively. Additionally, the Type-I error rate (E1) achieved exceptional results, with values of 0.0014, 0.0067, 0.0016, 0.0012, 0.0027, 0.0023, 0.0015, 0.0019, 0.0014, 0.0015, 0.0020, and 0.0069 across these datasets, further emphasizing the superiority of the proposed approach. Code is available at https://github.com/GeetanjaliGTZ/FIS-Net-NIR-and-VIS-Iris-Segmentation.
- 10 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 101472-101490
Open AccessFISNET: A Learnable Fusion-Based Iris Segmentation Network Improving Robustness Across NIR and VIS Modalities
Iris is one of the most distinctive biometric traits used for reliable identity verification in applications such as border control, secure access systems, and national ID programs. The primary challenge in iris recognition is the reliable segmentation of the iris region from the eye image. Segmenting irises captured under different imaging conditions is challenging for a single model due to variations in spectral features, texture, lighting conditions, and noise patterns between NIR and VIS images. To tackle this problem, we present Fused Iris Segmentation Network (FISNET) that combines segmentation maps from two models to achieve enhanced precision and accuracy. FISNET demonstrates robust generalizability across varying lighting, resolution, and sensor types, consistently outperforming individual models on all NIR and VIS datasets, including smartphone-captured images. The performance of FISNET was evaluated on CASIA-V4 subsets, Dark and Blue iris datasets, and the AFHIRIS dataset, achieving superior segmentation accuracy and recognition performance. The results demonstrate significant improvements over the IrisParseNet,PixlSegNet, and SAM model, achieving remarkable mIoU scores of 0.955, 0.930, 0.945, 0.955, 0.907, 0.815, 0.924, 0.913, 0.842, 0.852, 0.829, and 0.839 on the Lamp-V4, Interval-V4, Thousand-V4, Syn-V4, Twins-V4, UBIRIS-V2, BI-P1, BI-P2, DI-P1, DI-P2, DI-P3, and AFHIRIS-V1 datasets, respectively. Additionally, the Type-I error rate (E1) achieved exceptional results, with values of 0.0014, 0.0067, 0.0016, 0.0012, 0.0027, 0.0023, 0.0015, 0.0019, 0.0014, 0.0015, 0.0020, and 0.0069 across these datasets, further emphasizing the superiority of the proposed approach. Code is available at https://github.com/GeetanjaliGTZ/FIS-Net-NIR-and-VIS-Iris-Segmentation.
- Open Access
EAD: An EEG Adapter for Automated Classification
While electroencephalography (EEG) has been a popular modality for neural decoding, it often involves task specific acquisition of the EEG data. This poses challenges for the development of a unified pipeline to learn embeddings for various EEG signal classification, which is often involved in various decoding tasks. Traditionally, EEG classification involves the step of signal preprocessing and the use of deep learning techniques, which are highly dependent on the number of EEG channels in each sample. However, the same pipeline cannot be applied even if the EEG data is collected for the same experiment but with different acquisition devices. This necessitates the development of a framework for learning EEG embeddings, which could be highly beneficial for tasks involving multiple EEG samples for the same task but with varying numbers of EEG channels. In this work, we propose EEG Adapter (EAD), a flexible framework compatible with any signal acquisition device. More specifically, we leverage a recent EEG foundational model with significant adaptations to learn robust representations from the EEG data for the classification task. We evaluate EAD on two publicly available datasets achieving state-of-the-art accuracies 99.33% and 92.31% on EEG-ImageNet and BrainLat respectively. This illustrates the effectiveness of the proposed framework across diverse EEG datasets containing two different perception tasks: stimulus and resting-state EEG signals. We also perform zero-shot EEG classification on EEG-ImageNet task to demonstrate the generalization capability of the proposed approach.
Open AccessEAD: An EEG Adapter for Automated Classification
While electroencephalography (EEG) has been a popular modality for neural decoding, it often involves task specific acquisition of the EEG data. This poses challenges for the development of a unified pipeline to learn embeddings for various EEG signal classification, which is often involved in various decoding tasks. Traditionally, EEG classification involves the step of signal preprocessing and the use of deep learning techniques, which are highly dependent on the number of EEG channels in each sample. However, the same pipeline cannot be applied even if the EEG data is collected for the same experiment but with different acquisition devices. This necessitates the development of a framework for learning EEG embeddings, which could be highly beneficial for tasks involving multiple EEG samples for the same task but with varying numbers of EEG channels. In this work, we propose EEG Adapter (EAD), a flexible framework compatible with any signal acquisition device. More specifically, we leverage a recent EEG foundational model with significant adaptations to learn robust representations from the EEG data for the classification task. We evaluate EAD on two publicly available datasets achieving state-of-the-art accuracies 99.33% and 92.31% on EEG-ImageNet and BrainLat respectively. This illustrates the effectiveness of the proposed framework across diverse EEG datasets containing two different perception tasks: stimulus and resting-state EEG signals. We also perform zero-shot EEG classification on EEG-ImageNet task to demonstrate the generalization capability of the proposed approach.
- 4 April 2025
- Published by Springer Nature in International Journal on Document Analysis and Recognition (IJDAR)
- Vol. 28 (4) , 701-715
M2SM: multi modal signature matching network utilizing spatio-temporal features extracted from online signature
- 4 April 2025
- Published by Springer Nature in International Journal on Document Analysis and Recognition (IJDAR)
- Vol. 28 (4) , 701-715
M2SM: multi modal signature matching network utilizing spatio-temporal features extracted from online signature