Faculty Publications
Permanent URI for this communityhttps://idr.nitk.ac.in/handle/123456789/18736
Publications by NITK Faculty
Browse
4 results
Search Results
Item Word Sense Disambiguation using Bidirectional LSTM(Institute of Electrical and Electronics Engineers Inc., 2019) Rakshith, J.; Savasere, S.; Ramachandran, A.; Akhila, P.; Koolagudi, S.G.Word Sense Disambiguation is considered one of the challenging problems in natural language processing(NLP). LSTM-based Word Sense Disambiguation techniques have been shown effective through experiments. Models have been proposed before that employed LSTM to achieve state-of-the-art results. This paper presents an implementation and analysis of a Bidirectional LSTM model using openly available datasets (Semcor, MASC, SensEval-2 and SensEval-3) and knowledge base (WordNet). Our experiments showed that a similar state of the art results could be obtained with much less data or without external resources like knowledge graphs and parts of speech tagging. © 2019 IEEE.Item Hand classification based on fingerprint using Lightweight Convolutional Neural Network(Institute of Electrical and Electronics Engineers Inc., 2025) Akhila, P.; Koolagudi, S.G.Fingerprints are recognized as one of the most distinctive and reliable biometric identifiers that play a crucial role in forensic investigations by aiding in the swift identification of individuals. While traditional fingerprint analysis focuses on individual identification, determining the hand from which a particular fingerprint originates holds significant untapped potential. This paper proposes lightweight Convolutional Neural Networks to identify the hand from fingerprints. The model could achieve high accuracy on publicly available fingerprint datasets such as CASIA, SOCOFing, and NISTSD4. An in-depth analysis of the network prediction is conducted to determine the features that help the model identify the hand from the fingerprint. It is found that the position of core point, direction of ridge flow, inter-ridge distance at side ridges, and the slope of the ridges help the model identify the hand from fingerprints. © 2025 IEEE.Item Latent fingerprint segmentation using multi-scale attention U-Net(Inderscience Publishers, 2024) Akhila, P.; Koolagudi, S.G.Latent fingerprints are the fingerprints lifted from crime scene surfaces. Segmentation of latent fingerprints from the background is an important preprocessing task which is challenging due to the poor quality of the fingerprints. Though fingerprint segmentation approaches based on their orientation and frequency are reported in the literature, they could not adequately address the problem. We propose a latent fingerprint segmentation model based on the U-Net attention network in this work. We added the Atrous Spatial Pyramid Pooling (ASPP) layer to the network to facilitate multi-scale fingerprint segmentation. Our approach could effectively segment the latent fingerprint region from the background and even detect occluded and partial fingerprints with simple network architecture. To evaluate the performance, we have compared our results with the manual ground truth using NIST SD27A dataset. Our segmentation model has improved matching accuracy on the NIST SD27A dataset. © 2024 Inderscience Enterprises Ltd.Item End-to-end latent fingerprint enhancement using multi-scale Generative Adversarial Network(Elsevier B.V., 2024) Pramukha, R.N.; Akhila, P.; Koolagudi, S.G.Latent fingerprint enhancement is paramount as it dramatically influences matching accuracy. This process is often challenging due to varying structured noise and background patterns. The prints may be of arbitrary sizes and scales with a high degree of occlusion. There is a need for creating an end-to-end system that handles different conditions reliably to streamline this often lengthy and tricky process. In this work, we propose a Generative Adversarial Network (GAN) based architecture that effectively captures multi-scale context using Atrous Spatial Pyramid Pooling (ASPP). We have trained the network on a synthetically generated dataset, carefully designed to represent real-world latent prints. By avoiding the reconstruction of spurious ridges and only enhancing valid ridges, we avoid the generation of false minutiae, leading to better matching performance. We obtained state-of-the-art results in Sensor to Latent matching using the IIITD MOLF and Latent to Latent Matching using IIITD Latent datasets. © 2024 Elsevier B.V.