A comparison between different segmentation methods

Abstract

Upland heather moors are of vital importance in terms of economic and aesthetic values. A satisfying segmentation method is required to help the following classification thus helping land managers to make decisions about their management. This project focuses on object-based segmentation and four approaches: multiresolution segmentation of eCognition Professional 4.0, along with edge-based segmentation, marker controlled segmentation and efficient graph-based segmentation of python. They were investigated in terms of their performances and a comparison was made between them. Results showed that the quality of multiresolution segmentation is the best of the four. Edge-based segmentation result was limited to the sigma value that is used to smooth the image. The result of marker controlled segmentation was good to segment a certain aged heather stands. This is due to the limited number of user-defined markers. Efficient graph-based segmentation showed the closest result to that of multiresolution segmentation of eCognition Professional 4.0. In terms of the processing time, eCognition multiresolution segmentation took around 28 seconds, which is the longest. For edge-based segmentation, the time consumed was about 13 seconds, followed by efficient graph-based (16 seconds) and marker controlled segmentation (26 seconds). Therefore, python segmentation showed a potential to produce satisfying result with less processing time although a large amount of future work needs to be done.