In this work a new compression method for multispectral images has been proposed: the ‘colorimetric–spectral clustering’. The basic idea arises from the well-known cluster analysis, a multivariate analysis which finds the natural links between objects grouping them into clusters. In the colorimetric–spectral clustering compression method, the objects are the spectral reflectance factors of the multispectral images that are grouped into clusters on the basis of their colour difference. In particular two spectra can belong to the same cluster only if their colour difference is lower than a threshold fixed before starting the compression procedure. The performance of the colorimetric–spectral clustering has been compared to the k-means cluster analysis, in which the Euclidean distance between spectra is considered, to the principal component analysis and to the LabPQR method. The colorimetric–spectral clustering is able to preserve both the spectral and the colorimetric information of a multispectral image, allowing this information to be reproduced for all pixels of the image.
Colorimetric-Spectral Clustering: a tool for multispectral image compression
CIPRIAN, Roberta;
2011-01-01
Abstract
In this work a new compression method for multispectral images has been proposed: the ‘colorimetric–spectral clustering’. The basic idea arises from the well-known cluster analysis, a multivariate analysis which finds the natural links between objects grouping them into clusters. In the colorimetric–spectral clustering compression method, the objects are the spectral reflectance factors of the multispectral images that are grouped into clusters on the basis of their colour difference. In particular two spectra can belong to the same cluster only if their colour difference is lower than a threshold fixed before starting the compression procedure. The performance of the colorimetric–spectral clustering has been compared to the k-means cluster analysis, in which the Euclidean distance between spectra is considered, to the principal component analysis and to the LabPQR method. The colorimetric–spectral clustering is able to preserve both the spectral and the colorimetric information of a multispectral image, allowing this information to be reproduced for all pixels of the image.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
