Hierarchical cluster analysis methods applied to image segmentation by watershed merging

Jakub Smołka

Abstract


A drawback of watershed transformation is over-segmentation. It consists in creating moreclasses than there are objects present in the image. Over-segmentation partially results from thefact that the transformation extracts almost all edges present in the image, even those which arevery weak. To alleviate this problem images are preprocessed: blurring (or selectively blurring)filter is applied before the edge detection performed by a gradient filter. Additionally, the resultingimage may be thresholded in order to eliminate small gradient values.This paper presents an alternative solution to this problem. The solution uses the hierarchicalcluster analysis methods for joining similar classes of the over-segmented image into a givennumber of clusters. First, it calculates attribute values for each class. Second optionally, the valuesare standardized. Third, cluster analysis is performed. The resulting similarity hierarchy allows forsimple selection of the number of clusters in the final segmentation.Several clustering methods, including the Complete Linkage and Ward's method along withmany similarity/dissimilarity measures have been tested. The selected results are presented.

Full Text:

PDF


DOI: http://dx.doi.org/10.17951/ai.2007.6.1.73-84
Date of publication: 2015-01-04 00:00:00
Date of submission: 2016-04-27 10:20:01


Statistics


Total abstract view - 331
Downloads (from 2020-06-17) - PDF - 0

Indicators



Refbacks

  • There are currently no refbacks.


Copyright (c) 2015 Annales UMCS Sectio AI Informatica

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 International License.