Color quantization or color reduction is an important technique for image analysis and has been widely used in image segmentation

image compression and image recognition. Firstly

on bisecting K-means is used to quantize image roughly and then we refine the image by improved spectral clustering based weighted distance. The stability and quickness of bisecting K-means and adjustable weight make our approach an attractive one. Experimental results show that our approach performs better than octree algorithm in quantized quality and has a less computation complexity than K-means algorithm. For special image

which includes one important color but with only a few pixels

traditional approaches usually lose the important color

but our approach can deal with it by introducing the weight for distances between pixels.