1. Aggregation of Convolutional Neural Network Estimations of Homographies by Color Transformations of the Inputs
- Author
-
David Elizondo, Miguel A. Molina-Cabello, Ezequiel López-Rubio, Rafael Marcos Luque-Baena, [Molina-Cabello, Miguel A.] Univ Malaga, Dept Comp Languages & Comp Sci, Malaga 29071, Spain, [Marcos Luque-Baena, Rafael] Univ Malaga, Dept Comp Languages & Comp Sci, Malaga 29071, Spain, [Lopez-Rubio, Ezequiel] Univ Malaga, Dept Comp Languages & Comp Sci, Malaga 29071, Spain, [Molina-Cabello, Miguel A.] Biomed Res Inst Malaga IBIMA, Malaga 29010, Spain, [Marcos Luque-Baena, Rafael] Biomed Res Inst Malaga IBIMA, Malaga 29010, Spain, [Lopez-Rubio, Ezequiel] Biomed Res Inst Malaga IBIMA, Malaga 29010, Spain, [Elizondo, David A.] De Montfort Univ, Dept Comp Technol, Leicester LE1 9BH, Leics, England, Ministry of Economy and Competitiveness of Spain, Ministry of Science, Innovation and Universities of Spain, Autonomous Government of Andalusia, Spain, through the Project Detection of Anomalous Behavior Agents by the Deep Learning in Low Cost Video Surveillance Intelligent Systems, and European Regional Development Fund (ERDF)
- Subjects
General Computer Science ,Computer science ,business.industry ,General Engineering ,ComputingMethodologies_IMAGEPROCESSINGANDCOMPUTERVISION ,Pattern recognition ,Convolutional neural network ,Algorithm ,homography estimation ,Deep convolutional neural networks ,General Materials Science ,Deep homography ,Artificial intelligence ,lcsh:Electrical engineering. Electronics. Nuclear engineering ,business ,color transformations ,lcsh:TK1-9971 ,Sports - Abstract
The standard approach to the estimation of homographies consists in the application of the RANSAC algorithm to a set of tentative matches. More recent strategies based on deep learning, namely convolutional architectures, have become available. In this work, a new algorithm for the estimation of homographies is developed. It is rooted in a convolutional neural network for homography estimation, which is provided with a range of versions of the input pair of pictures. Such versions are generated by perturbation of the color levels of the input images. Each generated pair of images yields a distinct estimation of the homography, and then the estimations are combined together to obtain a final, more robust estimation. Experiments have been designed and carried out to test the validity of our approach, including qualitative and quantitative performance measures. In particular, it is demonstrated that our approach consistently outperforms the baseline approach consisting of using the output of the homography estimation deep network for the original input pair of images.
- Published
- 2020