Unsupervised classification by Isodata using genetic algorithm and Xie - Beni criterion

摘要

The unsupervised classification by the Isodata algorithm is closely dependent on the two parameters: the threshold to divide one class and the other threshold to merge two classes. Poor optimization of these two parameters leads the algorithm to escape any control retaining only one class in the end. The objective of this work is to make improvements to this algorithm; we used genetic algorithms (GA). We have designed a GA that will estimate the two optimal thresholds. These two parameters thus found will then be used by the Isodata algorithm. This approach is tested on simulation examples. The experimental results obtained confirm favorably the good performances of the proposed algorithm.