Index Geophysics

Line-based image segmentation method: a new approach to segment VHSR remote sensing images automatically

There exist different approaches for segmenting Very High Spatial Resolution (VHSR) remote sensing imagery with competitive performance, including object-based (e.g. Multiresolution), gradient-based (e.g. Watershed), and clustering-based (e.g. k-means) segmentation. However, they have a strong dependence on human assistance for tuning the required parameters (e.g. scale value, clusters number or tolerance thresholds), usually following a trial-and-error methodology that becomes tedious, hardly reproducible or transferable to other images, affecting negatively the methods’ robustness and efficiency. In this communication, we propose a novel method denominated Line-based segmentation (LBS) that automatically segments VHSR remote sensing imagery through a data-driven approach, bypassing the parameters’ definition by experts (i.e. region growing´s seeds and thresholds). The proposed algorithm offers flexibility and accuracy to segment regions with varying sizes and shapes, tested on different VHSR images, including multispectral images (WorldView-3, GeoEywe-1, Ikonos, QuickBird and SkySat), RGB aerial image (NAIP) and panchromatic image (Ikonos). The results revealed the LBS method shows a competitive performance compared against two well-known segmentation approaches, but without user intervention and generating consistent and repeatable segmentation results following an automatic fashion.

Jaime Lopez

John W. Branch

Gang Chen

image segmentation

remote sensing

automatic methodology

Oceanography

GC1-1581

Geology

QE1-996.5

Taylor & Francis Group

2019-01-01T00:00:00Z

article

2279-7254

10.1080/22797254.2019.1699449

https://doaj.org/article/51409d6453414858aa3d571a8e425b09

European Journal of Remote Sensing, Vol 52, Iss 1, Pp 613-631 (2019)

EN

http://dx.doi.org/10.1080/22797254.2019.1699449

https://doaj.org/toc/2279-7254

Journal Licence: CC BY-NC