Improved detection and clustering of hyperspectral image data by preprocessing with a euclidean distance transformation

Accurate detection and clustering are two of the main analysis tasks for remotely sensed spectral imagery. Hyper-spectral image (HSI) analysis often involves mathematically transforming the raw data into a new space using Principal Components Analysis (PCA) or similar techniques where a lower dimensional subspace containing most of the image information may be extracted. The results of standard algorithms may perform better in this new, less correlated space. Many of the currently used transformations in HSI analysis are statistical in nature and therefore place Gaussian or similar assumptions on the data distribution. A new, data driven, mathematical transformation is presented as a preprocessing step for HSI analysis. Termed the Nearest Neighbor Transformation, this new transformation does not rely on placing assumptions on the data. Instead, the approach taken is to use the pair-wise Euclidean distance between neigboring pixels in order to characterize the distribution of the data. This approach is introduced here and shown to improve analytical results from standard HSI algorithms, including anomaly detection and clustering.