Institutional Digital Repository (Indian Institute of Technology Kharagpu)

Majumder, Ranabir

Anticancer Activity of Aloe vera Leaves Extract against Human Breast Cancer: A Computational and Experimental Approach

Priyadarshini, Monali

Removal of Emerging Contaminants and Dyes from Wastewater Using Advanced Oxidation and Bio-Electrochemical Systems

Das, Tulika

Structure Based Mechanistic Insights into the Unwinding and Strand Annealing Activities of Human RECQ1 Helicase

Arohi, Adya Charan

Processing Induced Microstructural Evolution and Property Modification for Metastable β Ti-5Al-5V-5Mo-3Cr Alloy

Govindrao, Kale Mandar

Content based image retrieval using fuzzy class
membership for color images
Kale Mandar Govindrao
(14EC91P03)
Department of Electronics & Electrical Communication Engineering, Indian
Institute of Technology Kharagpur, India
Abstract
The rapid growth of internet users across the globe and extensive use of digital multimedia
devices have witnessed an exponential rise in imagery and visual information over the past decade.
In recent times, the daily use of digitized images has become an inseparable part of human life in
many fields such as entertainment, healthcare, education, e-commerce, etc. It has made information
retrieval techniques very relevant in the current era. The content-based image retrieval (CBIR)
systems exploit the visual information present in the images to find a suitable match similar to the
human visual saliency mechanism.
This thesis proposes an efficient CBIR system to retrieve color images. Most of the CBIR
systems are based on traditional distance-based retrieval. The thesis aims to improve retrieval ac-
curacy by utilizing the fuzzy class membership-based retrieval framework. First, a few significant
advancements are suggested to enhance the retrieval accuracy of the approach developed for tex-
tures. Next, the framework is generalized for natural color images and different classifiers. The set
of standard color descriptors are utilized for the retrieval. In addition, the high-level description
from pre-trained deep residual networks (ResNet) is explored for the improvements.
Although multiple descriptors are encouraging, they often have a high dimensionality that in-
creases computational overhead. The feature selection methods could help reduce the size by find-
ing the optimal set of features. However, it is difficult to decide which feature to try on before
the retrieval. Hence, we propose the deep stack autoencoder sparse autoencoder (DSSAE) and
convolutional autoencoder (CAE) to learn high-level descriptions directly from the images. DSSAE
and CAE models achieve a high level of abstraction and competing retrieval performance.
Next, DSSAE model further explores the high-level description of ResNet18 and ResNet50
networks in place of image data. The experiment highlights its efficacy in learning low-dimensional
latent presentation. The DSSAE model achieves a 50% reduction in the size of high-level description
while maintaining the highly competing retrieval performance. In addition, the thesis highlights
the impact of the database size on learning meaningful information using the DSSAE models. In
summary, the thesis presents an effective retrieval method for natural color images.
Keywords : Content-based image retrieval, color images, feature extraction, classification,
learning-based retrieval, fuzzy class membership, residual networks, sparse autoencoder,
convolutional autoencoder
1


Content Based Image Retrieval using Fuzzy Class Membership for Color Images

Mondal, Snigdha

India’s Renaissance into Bhārata: Examining nation, dharma and
literature in the light of select works of Sri Aurobindo
Snigdha Mondal
Roll no.-18HS92F12
Abstract
Article 1 of the Constitution of India declares, ‘India, that is Bharat shall be the union of states’
which while reinforcing the duality of its identity, also underlines the vital role that a ‘name’
plays in constructing and reconstructing a national identity (1950, p.1). References from ancient
times substantiate the fact that the concept of ‘India’ existed in various names and forms-whether
as bhāratas in the Rgveda (2000 BCE) or as Bhārata Shakti in Sri Aurobindo’s The Renaissance
in India and other essays on Indian culture (1997). Sri Aurobindo, one of the torchbearers of the
Indian nationalist movement, wrote the above series of essays, in defence of Indian culture and
civilisation, when the national identity of a people was suppressed under colonial rule leading to
the need for redefining its national ethos or svadharma that is encapsulated in the concept of
dharma. The efforts made by Sri Aurobindo and his co-revolutionaries in the beginning of the
twentieth century to reclaim their oppressed national identity during the colonial era included the
upholding of the literature of the land because texts like the Veda and the epics, viz. the
Rāmāyaṇa and the Mahābhārata played and continue to play a seminal role in infusing the
bhāratīya psyche with dhārmik principles. This thesis attempts to position India’s renaissance as
Bhārata in the contemporary scenario when there is a reawakening of national identity through
the restoration of the dharmic essence of the land and the role that literature can play in it based
on select writings of Sri Aurobindo where he expounds on the concepts and relationships of
nation, dharma and literature. This study follows a qualitative research methodology including
scoping method, thematic, narrative and content analysis for achieving its objectives to establish
the above.
Key words: India, Bhārata, Sri Aurobindo, nation, dharma, literature, renaissance


India’s Renaissance into Bhārata: Examining Nation, Dharma and Literature in the Light of Select Works of Sri Aurobindo

Goswami, Susmita

Groundwater-Thermal Water Characteristics and Interactions in the Coastal aquifers of Odisha, India

Palo, Patitapaban

Geometric Transformations and Learning Based Algorithms for Seismic Data Analysis and Interpretation

Das, Priyanka

Improvement of Pedestrian Infrastructure and Facilities at Tourist Destinations

Mukherjee, Rajdeep

Domain-Specific Text Summarization: Modeling and Evaluation

Not a member yet