Abstract: In order to solve the complicated and inefficient problem for organizing massive digital artistic images with general methods, this paper proposes a self-organization method based on visual similarity computation. For visual similarity computing, features of image such as color, texture, space layout and SIFT are extracted. According to the calculation model designed from the spatial layout of artistic images, the method calculates the clustering effect of images under different features, and adopts Multi-Layered version of Affinity Propagation （MLAP） clustering algorithm for various levels on given image database, by which it constructs a hierarchical structure on the visual information for these images. Experimental results show that the proposed method can achieve better organizational efficiency for artistic images.

Key words: artistic image, feature extraction, similarity computation, hierarchical clustering, self-organization, image management

摘要： 为解决大量数字化艺术图像常规组织和管理复杂低效问题，提出一种基于图像相似性计算的自组织方法，对艺术图像提取了颜色、纹理、空间布局和SIFT等用于相似性计算的视觉特征表示，并根据艺术图像空间布局特点设计计算模型，试验了特征的聚类效果。采用多层版本近邻传播聚类（MLAP）算法为基础，对实验图像库进行层次化聚类，构建图像的层次化浏览结构。实验结果表明，该方法在艺术图像的管理和使用上都有着良好的性能。

关键词: 艺术图像, 特征提取, 相似性计算, 层次聚类, 自组织, 图像管理