Residual Clustering Based Lossless Compression for Remotely Sensed Images

In the K-means algorithm, every pixel in a super-space is required to calculate Euclidean distance for clustering, so it is time-consuming computing when there are a great many class centers. Improved K-means clustering algorithm presented here could save initial clustering time by making initial division based on previous clustering results, and maintain the relationship among stable classes. Only calculating and comparing distances with neighbor centers, near to the pixel except those far away from it, accelerates clustering process with more and more classes becoming stable. Clustering lossless compression algorithm can efficiently eliminate the inter-spectral and intra-spectral redundancy at high convergent speed through enhancing intra-class redundancy. The multi-level clustering process can not only remove the spatial redundancy but also delete the residue redundancy, whose importance in lossless compression was overlooked previously, realizing a breakthrough lossless compression ratio at 2.882 for multi-spectral images. The comparison of the parameter analysis of the TM (Landsat Thematic Mapper) images with other lossless compression algorithms shows that this multilevel clustering lossless compression algorithm is more efficient.