K-Means Based Spatial Aggregation for Hyperspectral Compression

Summary form only given. We investigate a method to improve the compression ratio for hyperspectral data compression by use of a pre-processing step that gathers together correlated pixels before the transform is applied in a KLT-JPEG2000 based compression. Using a k-means clustering algorithm, the pixels can be grouped together before the application of the transform. Some similar methods have been studied, but k-means has been avoided due to its computational complexity. We call our proposed method SAMLC (Spatially Aggregated Multilevel Clustering). The simulation results show that in the case of lossy modes of compression, the proposed algorithm outperforms KLT+JPEG2000 and basic multilevel clustering for Hyperion imagery and some AVIRIS imagery. In lossless mode, Hyperion, AVIRIS, and AIRS data was tested but the proposed algorithm performed nearly the same as the competing algorithms across the 10 images tested. Overall, the proposed SAMLC algorithm is designed for lossy-to-lossless compression and performed best in lossy mode with Hyperion data.