提出了一种基于免疫谱聚类的图像分割方法.利用谱聚类的维数缩减特性获得数据在映射空间的分布,在此基础上构造一种新的免疫克隆聚类,用于在映射空间中对样本进行聚类.该方法通过谱映射为后续的免疫克隆聚类提供低维而紧致的输入.而免疫克隆聚类算法具有快速收敛到全局最优并且对初始化不敏感的特性,从而可以获得良好的聚类结果.在将其用于图像分割时,采用了Nystr?m逼近策略来降低算法复杂度.合成纹理图像和SAR图像的分割结果验证了免疫谱聚类算法用于图像分割的有效性.

An image segmentation approach based on immune spectral clustering algorithm, is proposed, in which the dimension reduction ability of the spectral clustering is used to attain the distribution of data in the mapping space. Next, a new immune clonal clustering algorithm is proposed to cluster the sample points in the mapping space. Compact input with low-dimension for immune clonal clustering is obtained after spectral mapping, and the immune clonal clustering algorithm, characterized by its rapid convergence to global optimum and minimal sensitivity to initialization, can obtain good clustering results. To efficiently apply the algorithm to image segmentation, Nystr?m method is used to reduce the computation complexity. Experimental results on synthetic texture images and SAR images show the validity of the algorithm in image segmentation.

Supported by the National Natural Science Foundation of China under Grant Nos.60672126, 60803097, 60703107, 60703108 (国家自然科学基金); the National Basic Research Program of China under Grant No.2006CB705700 (国家重点基础研究发展计划(973)); the National High-Tech Research