Firstly, on bisecting K-means is used to quantize image roughly and then we refine the image by improved spectral clustering based weighted distance.

首先利用高效的二分K均值聚类进行粗略量化，然后使用基于加权距离的谱聚类进行再次量化。