https://irju.jdvu.ac.in/jspui/handle/123456789/2 Wed, 04 Nov 2026 22:11:08 GMT 2026-11-04T22:11:08Z https://irju.jdvu.ac.in/jspui/handle/123456789/10811 Title: Bayesian edge-texture feature descriptor for human face recognition Authors: Sarkar, Abhijit Abstract: This thesis work describes a novel feature descriptor named Conditional Bayes Conjunctional Edge-Texture and Texture-Edge (CBCETTE) feature extractor for human face recognition. It is based on Euclidean inter-distances between the local centre of mass (LCOM) and pixels, which is proficient in creating an optimal feature set by combining edge and texture inclusive of invariant to illumination, rotation, and noise. Initially, the image is divided into blocks of size 3 × 3 and for each block, a local centre of mass (LCOM) is derived initially from the original image. Using this local centre of mass, from each block image an edge image and a texture image is extracted using a novel approach on the basis of the direction of light. Subsequently from the edge image and texture image, a probabilistic edge image and a probabilistic texture image are derived further to enunciate the combined edge texture feature set. Finally, Bayes’ conditional probabilistic approach is applied on both the probabilistic edge and the probabilistic texture image to obtain the probabilistic edge-texture and vice-versa feature set. The optimum feature set is the conjunction of the two probabilistic feature sets. This final optimum conjunction probabilistic feature set is finally used for classification purposes. It has been successfully tested on various challenging databases like Yale B face database, Extended Yale B face database, Yale face database, Faces95 database, and ORL database. It provides the accuracy of 99.38%, 98.77%, 97.58%, 97.92%, and 93.75% respectively on Yale B face database, Extended Yale B face database, Yale face database, Faces95 database, and ORL database that exhibits supremacy of the proposed strategy for human face recognition under various conditions. Mon, 01 Jan 2024 00:00:00 GMT https://irju.jdvu.ac.in/jspui/handle/123456789/10811 2024-01-01T00:00:00Z https://irju.jdvu.ac.in/jspui/handle/123456789/10810 Title: Probability density function based fusion of CNN models for face recognition from video Authors: Mazumdar, Saikat Abstract: This thesis describes Face recognition using CNN classifiers and final ranking for class predicted by the proposed method is done using rank-level fusion. This thesis works on multi instance of individual faces, multi algorithms based on probability density approaches and multiple classifiers. Fuzzy ranking is generated by outputs (confidence factors) of classifier using the density functions based formulation. VGG16, InceptionResnetV2 and Xception are used as CNN classifiers. For mathematical formulation used three different probability density functions are Gaussian Density Function, Triangle Density function and Trapezoidal Density function. Rank-level fusion is done to generate final ranking. Later, comparison between three density functions are implemented. The proposed method is successfully tested two datasets. One is Chokepoint dataset and another Custom dataset. The datasets are created by cropping faces from videos using MTCNN and OpenCV. The rank-level fusion is performing better than used three CNN models according to accuracy. The accuracy scores for Chokepoint dataset corresponding to Gaussian density function, Triangle Density function and Trapezoidal Density Function are 99.802%, 99.904% and 99.896% respectively and accuracy scores for Custom dataset are 99.762%, 99.716% and 99.75% respectively. It shows a supremacy regarding to face recognition. Mon, 01 Jan 2024 00:00:00 GMT https://irju.jdvu.ac.in/jspui/handle/123456789/10810 2024-01-01T00:00:00Z https://irju.jdvu.ac.in/jspui/handle/123456789/10809 Title: A comprehensive study on claim detection utilizing multi-task learning Authors: Das, Shankha Shubhra Abstract: This work delves into the dynamic domain of Natural Language Processing (NLP), which has experienced remarkable strides owing to the advent of machine learning and deep learning methodologies. Within this expansive field, the identification and assessment of claims within textual data have emerged as pivotal tasks, necessitating automated solutions to gauge their validity and relevance. Our exploration extends beyond the confines of English to encompass a spectrum of languages, with a keen focus on Indian linguistic diversity, thereby broadening the applicability of NLP techniques across diverse cultural and linguistic contexts. Central to our investigation is the paradigm of Multi-task Learning (MTL), which offers the promise of concurrent task execution to enhance overall model performance. While MTL has traditionally been applied to similar tasks, our research pioneers its application to dissimilar tasks, thereby pushing the boundaries of its utility and efficacy. To this end, we employ innovative fusion methodologies and a long-head approach in task-specific layer strategies, helping our MTL models with enhanced adaptability and accuracy. Furthermore, our study delves into the intricate domain of Span Identification, particularly honing in on Sentiment and Claim Span Identification tasks. Leveraging transfer learning techniques, we develop robust models capable of effectively extracting specific spans of text containing pertinent information across multiple languages. By imbuing our models with multilingual support, we strive to ensure their versatility and applicability in diverse linguistic contexts, particularly within the rich variety of the Indian linguistic landscape. In synthesizing these methodologies and approaches, our research contributes to the advancement of NLP applications, fostering greater understanding and automation of textual data analysis, especially within the intricate and diverse linguistic fabric of India. Mon, 01 Jan 2024 00:00:00 GMT https://irju.jdvu.ac.in/jspui/handle/123456789/10809 2024-01-01T00:00:00Z https://irju.jdvu.ac.in/jspui/handle/123456789/10808 Title: Medical image segmentation using novel u-net variants Authors: Halder, Akash Abstract: In the realm of medical imaging, precise segmentation holds the key to unlocking invaluable insights for diagnosis and treatment. Leveraging the power of deep learning, this thesis explores innovative methodologies to enhance the accuracy and efficiency of medical image segmentation. Recently, deep learning models, particularly U-Net-like architectures, have obtained notable success in medical image segmentation. Ensemble learning is a powerful approach that helps leverage the performance of an overall model by incorporating the decisions of multiple models. We propose an ensemble learning model, called BLUE-Net, using a combination of three models, namely the basic U-Net, Link-Net, and FPN. We combine the decisions of these models with a fuzzy-ranking scheme by using a modified Blumberg function to decide whether a pixel in an image is a part of a foreground pixel of the segmentation mask. We also propose a novel model, called SEAU-Net, which employs attention to the lower skip connections only, effectively integrating SE attention blocks within the skip connections of the encoder backbone. Highlevel semantic features are ensured by these attention-aided feature-maps from the lower layers to the higher layers in the decoder that are effectively utilized for the reconstruction of segmentation maps. Recognizing key features from input images is a potent strategy to enhance overall model performance. One way to achieve this is by integrating decisions from multiple outputs. Hence, we introduce the W2CU-Net, which is a dual decoder network based on Cosine Similarity, Weibull function and Weierstrass’ Elliptic Curve Equation. It is designed for liver and liver tumor segmentation from CT scan images. The model employs an ensemble scheme, which combines the decisions of two decoders using a modifiedWeierstrass’ elliptic curve equation while selecting important features using Cosine Similarity and Weibull function, to determine whether a pixel in a CT scan image belongs to the foreground of the segmentation mask. Three models are evaluated across six datasets: 3DIRCADb, CHAOS, LiTS for liver and liver tumor segmentation, BUSI for breast cancer segmentation, OCDC for oral cancer segmentation, and ISIC 2018 for skin cancer segmentation. Mon, 01 Jan 2024 00:00:00 GMT https://irju.jdvu.ac.in/jspui/handle/123456789/10808 2024-01-01T00:00:00Z