基于非局部信息和子空间的模糊C有序均值聚类的图像分割算法
Image Segmentation Algorithm Based on Non-Local Information and Subspace for Fuzzy C-Ordered Mean Clustering
-
摘要: 针对FCOM算法在处理含噪图像时, 由于没有考虑图像的空间信息. 因此, 不能取得理想的分割结果. 为了有效地去除噪声的干扰, 提出了一种基于非局部信息和子空间的模糊C有序均值聚类(Non-Local Information and Subspace for Fuzzy C-Ordered Mean, SFCOM_NLS)算法. 首先, 利用图像中给定的相似邻域结构的像素去提取当前像素的非局部空间信息. 其次, 计算每个像素的典型性, 并对其进行排序, 在每次迭代中更新像素的典型性, 提高像素聚类的准确性, 解决由于在聚类过程中存在相似类导致误分类问题. 最后, 引入子空间聚类概念, 为图像不同维度分配适当的权重, 提高彩色图像的分割性能. 通过含噪合成图像和彩色图像分割实验结果表明, 所提算法在模糊分割系数、模糊划分熵、分割精度和标准化互信息等性能方面均优于其他对比算法.Abstract: The FCOM algorithm does not consider spatial information of images when processing noisy images, which leads to suboptimal segmentation results. To effectively remove noise interference, the Non-Local Information and Subspace for Fuzzy C-Ordered Mean (SFCOM_NLS) algorithm is proposed. Firstly, the non-local spatial information of each pixel is extracted using the given similar neighborhood structure in the image. Secondly, the typicality of each pixel is calculated and sorted, and the typicality of pixels is updated in each iteration to improve the accuracy of pixel clustering and solve the misclassification problem caused by similar classes during the clustering process. Finally, the concept of subspace clustering is introduced to allocate appropriate weights to different dimensions of the image, thus improving the segmentation performance of color images. Experimental results on synthetic images with noise and color image segmentation show that the proposed algorithm outperforms other compared algorithms in terms of fuzzy segmentation coefficient, fuzzy partition entropy, segmentation accuracy, and normalized mutual information.