基于颜色的自组织聚类分析图象检索算法

本文提出了一种改进的迭代自组织聚类分析方法（ISODATA）来实现基于颜色的图象检索。算法选择符合的人的视觉的HSV模型，充分发挥了色度的描述作用。同时，选择HSV空间上的颜色直方图作为颜色特征，该特征描述了图像颜色的统计分布特征，且具有平移，尺度和旋转不变性，算法改进了ISODATA的初始聚类中心的选取方法，其初始聚类中心的选择是面向数据分布的，具有较好的通用性，因此适合各种分布特性的数据对象，实验结果表明，算法的检索精度明显好于传统的基于RGB直方图的检索结果。

Image Retrieval Based on Color by ISODATA Algorithm

By means of improved Iterative Self-Organizing Data Analysis Techniques Algorithm (ISODATA), a new algorithm for color-based image retrieval is proposed. This algorithm is based on HSV color model, which is in accordance with human's perception and discrimination concerning color. Initial cluster centers are intentionally chosen according to the data distribution, thus diversifying its applications. Moreover, both the heuristic adjustment for clustering results and the cut-and-try method as well as human-computer interaction improve the image retrieval more efficiently. Experiment results show that using this algorithm, color-based image retrieval can be facilitated more efficiently than original RGB histogram algorithm.