Integer Wavelet Transformation Enhanced Multi-level Clustering Lossless Compression Algorithm for Remotely Sensed Images

Efficient clustering is essential for compressing high-dimensional remote sensing data, yet the conventional K-means algorithm is computational-ly intensive, as it requires calculating the Euclidean distance between every pixel and numerous class centers. To address this limitation, we present an im-proved K-means algorithm that reduces initialization time by leveraging results from previous clustering iterations. This approach preserves stable inter-class relationships and accelerates convergence, as class assignments become in-creasingly stable through distance evaluations restricted to neighboring centers. Building on this foundation, we develop a clustering-based lossless compres-sion algorithm capable of automatically determining the optimal number of classes and achieving high compression ratios. By effectively reusing prior clustering results, the method eliminates both inter-spectral and intra-spectral redundancies while enhancing intra-class redundancy convergence. In addition, we introduce the concept of residue redundancy, redundancy among residual data, which has been largely overlooked in earlier studies. The proposed Integer Wavelet Transformation Enhanced Multi-level Clustering Lossless Compres-sion Algorithm (IWT-MCLCA) integrates multi-level clustering with an integer wavelet transformation, which removes not only spatial and structural redun-dancies but also residue redundancies, thereby achieving a breakthrough in loss-less compression for multispectral images. Experimental results on Landsat Thematic Mapper (TM) data demonstrate a compression ratio of up to 3.842. Comparative analysis further indicates that the proposed multilevel clustering compression algorithm outperforms existing lossless approaches in both effi-ciency and effectiveness.