Image segmentation is a key technique of image processing and computer vision field. However, facing with large amount of image segmentation methods, the qualitative and quantitative evaluation of algorithms is very significant. This paper states the thoughts of high resolution RS image segmentation methods evaluation and tests it by evaluating four typical image segmentation algorithms based on features with six images qualitatively and quantitatively. The four typical image segmentation algorithms are Max-Entropy, Split & Merge, modified Gauss Markov Random Field and Orientation&Phase based Filters. In the qualitative evaluation, this paper analyses these algorithms in term of their basic principles and gets a rough evaluation. In the quantitative evaluation, image complexity is taken into account firstly and six measures are employed. The six measures are removed region number, nonuniformity within region measure, contrast across region measure, variance contrast across region measure and edge gradient measure. The qualitatively and quantitatively evaluation results is important to perform the optimal selection of segmentation algorithm in practical work. In the end, this paper analyzes the defects of image segmentation evaluation methods proposed by this paper and indicates the application prospect of high resolution RS image segmentation.
[1]
M. Neubert,et al.
EVALUATION OF REMOTE SENSING IMAGE SEGMENTATION QUALITY – FURTHER RESULTS AND CONCEPTS
,
2006
.
[2]
Peng-Yeng Yin,et al.
Maximum entropy-based optimal threshold selection using deterministic reinforcement learning with controlled randomization
,
2002,
Signal Process..
[3]
Y. J. Zhang,et al.
A survey on evaluation methods for image segmentation
,
1996,
Pattern Recognit..
[4]
Zhanfeng Shen,et al.
Multi‐scale information extraction from high resolution remote sensing imagery and region partition methods based on GMRF–SVM
,
2007
.