Open AccessImage Georeferencing using Artificial Neural Network Compared with Classical Methods
Author(s) -
Zahra Ezz El Din
Publication year - 2021
Publication title -
iraqi journal of science
Language(s) - English
Resource type - Journals
eISSN - 2312-1637
pISSN - 0067-2904
DOI - 10.24996/ijs.2021.62.12.38
Subject(s) - georeference , computer science , affine transformation , artificial intelligence , photogrammetry , transformation (genetics) , artificial neural network , coordinate system , computer vision , matlab , spatial reference system , geographic coordinate conversion , remote sensing , geography , mathematics , geometry , biochemistry , chemistry , physical geography , gene , operating system
Georeferencing process is one of the most important prerequisites for various geomatics applications; for example, photogrammetry, laser scan analysis, remotely sensing, spatial and descriptive data collection, and others. Georeferencing mostly involves the transformation of coordinates obtained from images that are inhomogeneous due to accuracy differences. The georeferencing depends on image resolution and accuracy level of measurements of reference points ground coordinates. Accordingly, this study discusses the subject of coordinate’s transformation from the image to the global coordinates system (WGS84) to find a suitable method that provides more accurate results. In this study, the Artificial Neural Network (ANN) method was applied, in addition to several numerical methods, namely the Affine divided difference, Newton’s divided difference, and polynomial transformation. The four methods were modelled and coded using Matlab programming language based on an image captured from Google Earth. The image was used to determine reference points within the study area (University of Baghdad campus). The findings of this study showed that the ANN enhanced the results by about 50% in terms of accuracy and 90% in terms of homogeneity, compared with the other methods.