结合近邻传播聚类的自适应图像分割

Adaptive Image Segmentation Using Affinity Propagation Clustering

提出一种统一的图像自动分割模型。为了将图像分为颜色、纹理相近的不同的区域,提出了一个处理方法,具体分为两个步骤:首先,用改进的简单线性迭代聚类算法对输入图像进行预处理,即过分割;然后,用其低阶颜色矩表示这些区域的特征,并进一步利用近邻传播聚类算法将这些区域进行合并。在公开的数据集上进行了详细的实验,结果证明了所提算法的有效性和健壮性。

This paper presented a unified approach for automatic image segmentation.In order to segment the image into homogenous regions,a two-stage method was proposed.Firstly,an improved simple linear iterative clustering method is adopted for the over-segmentation of the image.Then,color moments of each local region are computed to represent the region,and the affinity propagation clustering is adopted to merge the regions which are segmented in the first stage.Numerous experiments were conducted on public available datasets to demonstrate the effectiveness and robustness of the proposed algorithm.