Data Rate Reduction. Volume 3: Cascaded Clustering of Compressed Multispectral Image Data

摘要

A cascaded, clustering strategy which provides a globally shorter description of digital pictorial data with respect to the number of different multispectral vectors contained in a given frame of imagery is proposed. The implementation of cascaded clustering was studied. Cascaded clustering as well as pictorial reconstruction results demonstrate that affecting cascaded clustering by a hierarchical method preserves the image dynamic range and subjective fidelity of reconstruction. Next, this form of preprocessing was validated for supervised classification. A maximum likelihood supervised classification was performed on multispectral data. The results were compared with those obtained from both the raw data and the compressed data. On cascaded cluster data or on compressed data, the classification results are better than on raw data.