28 October 2006 Shape-information-based multi-scale watershed transform
Author Affiliations +
Proceedings Volume 6421, Geoinformatics 2006: Geospatial Information Technology; 64210Q (2006) https://doi.org/10.1117/12.712900
Event: Geoinformatics 2006: GNSS and Integrated Geospatial Applications, 2006, Wuhan, China
The watershed transform is powerful for image segmentation and has been successfully utilized in many different computer vision applications. It can always produce a complete division of the image. However it does not take into account of the prior information about the segmentation task. Another drawback of the watershed transform is over- segment or under segment, which means that the input image is divided into too many meaningless small areas or divided into too few large areas. In this paper, the authors have proposed a method which enables the introduction of prior information in the segmenting process. The prior information about the extracted objects is used to carry out multi-scale watershed. The multi-scale watershed consists of two main parts. The first is to build the scale space. The second is to integrate the segment result at each scale with the help of prior information to obtain the final segmented result. The final result is better than the result obtained in each scale alone. The proposed approach was tested on remote sensing image, and the results are compared with other image segmenting methods.
© (2006) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Tiancan Mei, Tiancan Mei, Deren Li, Deren Li, Qianqing Qin, Qianqing Qin, } "Shape-information-based multi-scale watershed transform", Proc. SPIE 6421, Geoinformatics 2006: Geospatial Information Technology, 64210Q (28 October 2006); doi: 10.1117/12.712900; https://doi.org/10.1117/12.712900

Back to Top