Every pixel in the super space is required by K means algorithm to calculate Euclidean distance for clustering. When there are much many class centers

this is a much hard work. In this paper

an improved K means clustering algorithm is presented to accelerate clustering process with more and more classes becoming stable by judging with neighbor centers nearest to the pixel. The inter spectral redundancy and intra spectral redundancy can be eliminated mostly by multilevel clustering algorithm with quickly convergent K means classification and the method clearing redundancy at step through enhancing the intra class pixel redundancy. The multi level clustering process with initial S+P (Sequential transform + Prediction) integer wavelet transformation can not only remove the spatial and structural redundancy

but also delete the residual data redundancy realizing the breakthrough of lossless compression for multi spectral images. Furthermore

the comparison with other lossless compression algorithm and the parameter analysis of the TM (Landsat Thematic Mapper) images show that this multilevel clustering compression algorithm is more reasonable and efficient.