Indian Journal of Science and Technology
DOI: 10.17485/IJST/v16i42.1782
Year: 2023, Volume: 16, Issue: 42, Pages: 3803-3813
Original Article
Laukikkumar K Patel1,2*, Manish I Patel3
1Ph.D. Scholar, Sankalchand Patel University, Visnagar, 384315, Gujarat, India
2Lecturer, Department of Electronics and Communication Engineering, Government Polytechnic, Palanpur, Gujarat, India
3Assistant Professor, Department of Electronics and Communication Engineering, Institute of Technology, Nirma University, Ahmedabad, Gujarat, India
*Corresponding Author
Email: [email protected]
Received Date:18 July 2023, Accepted Date:03 October 2023, Published Date:13 November 2023
Objectives: To improve image registration by reducing the estimation error of the rotation transformation parameter under illumination change effect in remote sensing images using Oriented Fast and Rotated Brief (ORB) and Convolutional Neural Network (CNN). Also, to reduce computational complexity that can be increased due to use of CNN. Methods: The image registration process aligns two or more images geometrically and a novel feature based approaches for image registration is proposed here, where ORB and CNN are used to estimate rotation transformation parameter under illumination change effects in remote sensing images with novelty in generation of feature descriptor. The results of the proposed approach are compared with an approach which uses only ORB for both feature detection and descriptor generation. Findings: In the proposed approach, convolutional features from modified CNN are used in a novel manner with ORB descriptor to generate the final fusion descriptor. Also, this novel approach reduces the computational complexity by limiting the descriptor size that can be increased due to CNN. Here, three different combinations of CNN layers are provided for the generation of descriptor with ORB and this approach is also tested with transfer learning concept and other than remote sensing image, which shows improved results for taken cases. The results of novel approach show that the estimation of rotation transformation parameter and image registration is improved, and the estimation error is reduced to 0.1% to 0.9% for taken cases. Novelty: Novelty is provided in the generation of descriptor by fusion of CNN (modified visual geometry group (VGG19)) features with descriptor from ORB for reduction in estimation error, limiting descriptor size to reduce computational complexity, and improvement in image registration of remote sensing images. It is also tested for transfer learning case and other than remote sensing image, where improved results are also seen.
Keywords: Image registration, CNN, Remote sensing, VGG19, Oriented fast and rotated brief
© 2023 Patel & Patel. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Published By Indian Society for Education and Environment (iSee)
Subscribe now for latest articles and news.