Abstract: ISODATA algorithm is capable of splitting and merging categories automatically. However, this kind of hard clustering fails to take into consideration the characteristics of image itself and human visual features. So its effect is generally not as good as that of fuzzy clustering algorithm. For most fuzzy recognition methods, if they are to be applied, the number of categories must be set beforehand. Besides, there is inherent defect in traditional fuzzy algorithms. By contrast, intuitionistic fuzzy is a kind of improvement to make up for the deficiencies of traditional fuzzy theory. Based on the advantages of ISODATA algorithm and intuitionistic fuzzy, with those critical functions which are related to membership and non-membership functions used as the measurement for clustering, this paper proposes a kind of ISODATA algorithm that is based on intuitionistic fuzzy, and introduces membership function that has been improved for practical purposes. This kind of function takes region as the sample to be classified. This paper verifies the effectiveness of the proposed algorithm by applying it to image segmentation.

Key words: intuitionistic fuzzy, image segmentation, Iterative Self-Organizing Data Analysis Technique Algorithm（ISODATA）

摘要： ISODATA算法能自动地进行类的分裂和合并，但这种硬分类算法没有充分考虑图像本身的特点和人类的视觉特性，其分类效果一般差于模糊聚类算法。而大多数模糊识别方法都需要设置类别数目，有其自身的缺点，而直觉模糊则弥补了传统模糊理论不足。结合直觉模糊和ISODATA优点，将与隶属度和非隶属度相关的判定函数作为分类度量，提出了一种基于直觉模糊的ISODATA算法，结合实际改进了隶属度函数，以区域为待分类样本以提高算法速度，将其应用到图像分割，经实验证明了算法的有效性。

关键词: 直觉模糊, 图像分割, 迭代自组织数据分析技术算法（ISODATA）