Recognition of Single-Land Countries on Outline Images by Using BAS Feature

Abstract

This paper addresses an original problem. It presents an approach for the recognition of single-land countries on outline satellite images by using Beam Angle Statistics (BAS). For this purpose, we have created an image dataset. After converting the RGB input images to binary and applying the necessary preprocessing steps, a single contour is extracted from each image by Canny Edge detector. In order to represent each county as a one-dimensional vector, BAS feature is extracted from contour point vector. Thus, each country has its unique feature vector and is independent from rotation, scale and is also robust to image deformation. For comparison of the one-dimensional vectors, Dynamic Time Warping (DTW) is used. In experiments, we had Additive White Noise (AWN) added to test images which are scaled with factors 1.5, 0.5, 0.25 and are also iteratively rotated by 45 degrees. Benchmarking has been conducted between the proposed algorithm, Scale Invariant Feature Transform (SIFT) and Oriented Fast Rotated Brief (ORB) features. Results show that BAS feature is the most robust feature against rotation followed by ORB and then SIFT. In case of upscale, BAS outperforms the others. In downscale versions, SIFT outperforms others and ORB comes as last.