Please use this identifier to cite or link to this item:
http://hdl.handle.net/10603/324538
Title: | Content Based Image Retrieval Using Features from Deep Convolutional Neural Networks |
Researcher: | Bhandi Vijaykumar |
Guide(s): | Sumithra Devi K A |
Keywords: | Computer Science Computer Science Software Engineering Engineering and Technology |
University: | Jain University |
Completed Date: | 2020 |
Abstract: | Content based image retrieval (CBIR) system makes use of low-level visual features to measure newlineimage similarity during image search. Color, texture, and shape properties are well known newlinehandcrafted features applied in traditional CBIR systems. The selection of handcrafted features is newlinevery vital in designing CBIR applications. Prior domain knowledge of the application is very newlineimportant for choosing appropriate handcrafted features. Usage of inappropriate features widens newlinethe semantic gap and leads to poor results. Automatic feature extraction which is independent of newlinedomain understanding is very essential. Machine learning techniques enable systems to learn newlineimportant representations from input image data. Convolutional neural networks (CNNs) are a newlinespecialized implementation of machine learning techniques and are able to create expressive newlinerepresentations from the input image. Hence CNNs are well suited for image processing operations newlinesuch as classification, clustering, and object detection, etc. The creation of a new effectual deep newlineCNN model involves an extensive training phase. This requires very large datasets, huge newlinecomputation environments, and longer execution time. Several established deep CNNs are readily newlineavailable. These networks are pre-trained on massive databases of images. VGG, ResNet, and newlineInception are the leading pre-trained CNN models currently being used in numerous image newlineprocessing studies. Possibly we can transfer knowledge learned from such models in order to newlineaddress challenges in different domains. This can be achieved by repurposing a deep CNN model newlineas a feature generator to produce effective features for the CBIR application. In this research work, newlinewe proposed multiple CBIR methodologies that implement extraction of features by utilizing newlinedifferent pre-trained CNN models. All these proposed methods have undergone extensive newlineexperimentation on various public datasets taken from diverse domains, such as medical, newlineradiology, meteorology, satellite imagery, and phytomorphology. The outcome of the empirical newlinestudy demonstrates that the features extracted from deep CNN models are excellently suited for newlineCBIR tasks. The proposed CBIR methods outperform handcrafted features across various domains newlineselected in this study. newline |
Pagination: | 134 p. |
URI: | http://hdl.handle.net/10603/324538 |
Appears in Departments: | Department of Computer Science Engineering |
Files in This Item:
File | Description | Size | Format | |
---|---|---|---|---|
80_recommendation.pdf | Attached File | 304.56 kB | Adobe PDF | View/Open |
certificate.pdf | 40.63 kB | Adobe PDF | View/Open | |
chapter-1.pdf | 173.22 kB | Adobe PDF | View/Open | |
chapter-2.pdf | 351.8 kB | Adobe PDF | View/Open | |
chapter-3.pdf | 91.53 kB | Adobe PDF | View/Open | |
chapter-4.pdf | 659.95 kB | Adobe PDF | View/Open | |
chapter-5.pdf | 541.28 kB | Adobe PDF | View/Open | |
chapter-6.pdf | 536.4 kB | Adobe PDF | View/Open | |
chapter-7.pdf | 1.17 MB | Adobe PDF | View/Open | |
chapter-8.pdf | 690.32 kB | Adobe PDF | View/Open | |
chapter-9.pdf | 2.38 MB | Adobe PDF | View/Open | |
conclusion.pdf | 190.05 kB | Adobe PDF | View/Open | |
coverpage.pdf | 114.91 kB | Adobe PDF | View/Open | |
table-of-contents.pdf | 222.96 kB | Adobe PDF | View/Open |
Items in Shodhganga are licensed under Creative Commons Licence Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0).
Altmetric Badge: