Seamless contiguity method for parallel segmentation of remote sensing image
Lists
Only you see this
Abstract
Seamless contiguity is the key technology for parallel segmentation of remote sensing data with large quantities. It can be effectively integrate fragments of the parallel processing into reasonable results for subsequent processes. There are numerous methods reported in the literature for seamless contiguity, such as establishing buffer, area boundary merging and data sewing. et. We proposed a new method which was also based on building buffers. The seamless contiguity processes we adopt are based on the principle: ensuring the accuracy of the boundary, ensuring the correctness of topology. Firstly, block number is computed based on data processing ability, unlike establishing buffer on both sides of block line, buffer is established just on the right side and underside of the line. Each block of data is segmented respectively and then gets the segmentation objects and their label value. Secondly, choose one block(called master block) and do stitching on the adjacent blocks(called slave block), process the rest of the block in sequence. Through the above processing, topological relationship and boundaries of master block are guaranteed. Thirdly, if the master block polygons boundaries intersect with buffer boundary and the slave blocks polygons boundaries intersect with block line, we adopt certain rules to merge and trade-offs them. Fourthly, check the topology and boundary in the buffer area. Finally, a set of experiments were conducted and prove the feasibility of this method. This novel seamless contiguity algorithm provides an applicable and practical solution for efficient segmentation of massive remote sensing image.
BibTeX
@inproceedings{wang2015seamless,
author = {Geng Wang and Guanghui Wang and Mei Yu and Chengling Cui},
title = {{Seamless contiguity method for parallel segmentation of remote sensing image}},
booktitle = {Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE},
year = {2015},
doi = {10.1117/12.2209034},
}Loading related content...