An Unsupervised Segmentation Algorithm for Remote Sensing Images

Abstract : Geographic Object-Based Image Analysis (GEOBIA) provided the potential for exploiting geographical information, such as spatial context or object shape. Because of that, image segmentation became a fundamental and a necessary task on the GEOBIA processing chain, though that already was a traditional operation in Remote Sensing and a fundamental problem of image processing. Regardless of the good results obtained by some algorithms, the correct partitioning of an image is still an unsolved problem. In this paper, we propose a new unsupervised context-based segmentation algorithm, which uses some concepts of the Bayesian approach, to do image partition for high-resolution remote sensing images. We use well-known and simple techniques to create this algorithm, aiming to label disjoint regions with similar characteristics and to reduce the computational cost by simplifying the energy function, thus facilitating information fusion through long-range interaction. The algorithm was tested on a WorldView-2 multispectral image with 98 reference objects and its results were evaluated with 6 quality measures, that analyse precision, recall, f1-score, the agreement to size, shape and positional adjustment. In addition, we compared the results of the proposed algorithm with those produced by the Multiresolution Segmentation algorithm, which is commonly applied by the GEOBIA community, the Meanshift algorithm, and a Region Growing segmentation algorithm. For the metrics of precision, recall, f1-score and size, the Multiresolution obtained the best results, while our proposed algorithm outperformed the others, on the metrics of positional adjustment and shape agreement.
Complete list of metadatas

http://hal.univ-reunion.fr/hal-01960446
Contributor : Réunion Univ <>
Submitted on : Wednesday, December 19, 2018 - 2:00:29 PM
Last modification on : Monday, June 10, 2019 - 9:32:02 AM

Identifiers

  • HAL Id : hal-01960446, version 1

Collections

Citation

Anderson Reis Soares, Thales Sehn Körting, Leila Maria Garcia Fonseca. An Unsupervised Segmentation Algorithm for Remote Sensing Images. GEOBIA 2018 - From pixels to ecosystems and global sustainability ​, Jun 2018, Montpellier, France. ⟨https://geobia2018.sciencesconf.org/⟩. ⟨hal-01960446⟩

Share

Metrics

Record views

56