Unsupervised texture image segmentation using multiobjective evolutionary clustering ensemble algorithm
2008
Multiobjective evolutionary clustering approach has been successfully utilized in data clustering. In this paper, we propose a novel unsupervised machine learning algorithm namely multiobjective evolutionary clustering ensemble algorithm (MECEA) to perform the texture image segmentation. MECEA comprises two main phases. In the first phase, MECEA uses a multiobjective evolutionary clustering algorithm to optimize two complementary clustering objectives: one based on compactness in the same cluster, and the other based on connectedness of different clusters. The output of the first phase is a set of Pareto solutions, which correspond to different tradeoffs between two clustering objectives, and different numbers of clusters. In the second phase, we make use of the meta-clustering algorithm (MCLA) to combine all the Pareto solutions to get the final segmentation. The segmentation results are evaluated by comparing with three known algorithms: K-means, fuzzy K-means (FCM), and evolutionary clustering algorithm (ECA). It is shown that MECEA is an adaptive clustering algorithm, which outperforms the three algorithms in the experiments we carried out.
Keywords:
- Single-linkage clustering
- Fuzzy clustering
- Correlation clustering
- Artificial intelligence
- Machine learning
- k-medians clustering
- FLAME clustering
- Cluster analysis
- CURE data clustering algorithm
- Algorithm
- Canopy clustering algorithm
- Pattern recognition
- Mathematics
- Determining the number of clusters in a data set
- Computer science
- Data stream clustering
- Correction
- Source
- Cite
- Save
- Machine Reading By IdeaReader
19
References
13
Citations
NaN
KQI