Please use this identifier to cite or link to this item: https://idr.l2.nitk.ac.in/jspui/handle/123456789/17072
Full metadata record
DC FieldValueLanguage
dc.contributor.advisorP, Jidesh.-
dc.contributor.authorP, Febin I.-
dc.date.accessioned2022-02-01T10:48:33Z-
dc.date.available2022-02-01T10:48:33Z-
dc.date.issued2021-
dc.identifier.urihttp://idr.nitk.ac.in/jspui/handle/123456789/17072-
dc.description.abstractImage restoration and enhancement are the two inevitable pre-processing activities that we come across in almost all imaging applications. Apparently, these two requirements contradict each other as one is a complement of the other. The image restoration aims at smoothing out the signal to reduce the noise interventions. On the other hand, enhancement seeks for an image with non-smooth features. Therefore, one should aim for a trade-off between these two requirements when providing a solution. Perceptually inspired frameworks have taken a considerable lead in image restoration and enhancement activities, as they seek for a visually appealing solution in addition to their excellent performance in terms of statistical quantifications. Retinex framework is being explored extensively in the literature to provide the desired enhancement to the images under consideration. This thesis provides an in-depth insight into various restoration frameworks and contributes a set of state-of-the-art restoration and enhancement models to assist the preprocessing step of various imaging applications with specific relevance to satellite and medical imaging. The degradation analysis is the primary step in an automated restoration framework. As one cannot apply a blanket restoration model for all kinds of distortions, the appropriate models are designed in due respect to the noise distribution of the input data. The second chapter of the thesis contributes a fully automated framework for analysing and detecting the noise distribution of the noise from input data. Analysis of noise distribution duly provides an insight to choose appropriate variational model to restore the images from the specific degradation analysed therein. A machine learning approach is employed to analyse the noise distribution from the input image characteristics. Various statistical and geometric features of the images are analysed to arrive at the conclusion regarding the distribution. Subsequent to the noise distribution analysis, the respective retinex based variational models are chosen to restore and enhance the images. One of the major issues with the variational models is that, they converge slowly when explicit numerical schemes are used for solving them. Many models designed under this framework use the explicit schemes due to the ease of implementation. Fast numerical implementations are one of the requirements of a realtime application model. This thesis investigates some of the fast numerical schemes such as Bregman iteration scheme redesigned for the problem under consideration to effectively solve the problems. Moreover, the computational cost is a major matter for i concern among the scientists, as most of the practically viable systems should be computationally efficient to be used under a real-time scenario. This thesis addresses this issue considerably well by employing parallel computing algorithms designed to be executed under multi-processing environments to improve the computational efficiency of the model.en_US
dc.language.isoenen_US
dc.publisherNational Institute of Technology Karnataka, Surathkalen_US
dc.subjectDepartment of Mathematical and Computational Sciencesen_US
dc.subjectPerceptually inspired modelen_US
dc.subjectRetinex frameworken_US
dc.subjectVariational restoration modelsen_US
dc.subjectData-correlated noiseen_US
dc.subjectImage enhancementen_US
dc.subjectSatellite and medical image enhancementen_US
dc.titlePerceptually Inspired Variational Retinex Methods for Enhancing and Restoring Imagesen_US
dc.typeThesisen_US
Appears in Collections:1. Ph.D Theses

Files in This Item:
File Description SizeFormat 
FEBIN_THESIS_MACS.pdf18.65 MBAdobe PDFThumbnail
View/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.