一种改进的基于归一化割的图像分割方法

An Improved Approach for Image Segmentation Based on Normalized Cut

文章结合聚类法提出了一种改进的基于归化割的图像分割方法。先利用聚类对图像进行预分割,得到图像最大相似区域,再利用区域间的亲近关系将图像转化为图,最后利用图谱的工具(特征值和特征向量)对图分割,实现图像的分割。实验结果表明,该方法有效地降低算法复杂度,获得较好的分割效果。

An approach for image segmentation by Normalized Cut with clustering is proposed in this paper. Firstly, using clustering to convert images to graphs, then calculates the spectral graph properties(eigenvalues and eigenvectors) of the similarity matrix. Finally, partition the graph by the second smallest eigenvalue denotes the way to segment an image into meaningful regions. The experiment shows that the algorithm reduces the complexity effectively, and receives good results of segmentation.