A fuzzy quantizer of multidimensional vectors for color image compression is proposed. It exploits the spatial redundancy as well as the chromatic information by defining the membership function with spatial-chromatic combined vectors. As its hypersphere determined by the membership function shrinks

each vector

which is assigned to multiple clusters in the early stages of iteration

falls in one cluster. The initial codebook is generated by a constrained random initialization to help avoid the local minimum. Experiment results demonstrated that this approach of spatial-chromatic vector quantization can achieve high compression efficiency. The results also revealed that the method is capable of diminishing noise without blurring edges in the image.