gen_principal_comp_trans — Compute the transformation matrix of the principal component analysis of multichannel images.
gen_principal_comp_trans computes the transformation matrix of a principal components analysis of multichannel images. This is useful for images obtained, e.g., with the thematic mapper of the Landsat satellite. Because the spectral bands are highly correlated, it is desirable to transform them to uncorrelated images. This can be used to save storage, since the bands containing little information can be discarded, and with respect to a later classification step.
The operator gen_principal_comp_trans takes one or more multichannel images MultichannelImage and computes the transformation matrix Trans for the principal components analysis, as well as its inverse TransInv. All input images must have the same number of channels. The principal components analysis is performed based on the collection of data of all images. Hence, gen_principal_comp_trans facilitates using the statistics of multiple images.
If n is the number of channels, Trans and TransInv are matrices of dimension n × (n+1), which describe an affine transformation of the multichannel gray values. They can be used to transform a multichannel image with linear_trans_color. For information purposes, the mean gray value of the channels and the n × n covariance matrix of the channels are returned in Mean and Cov, respectively. The parameter InfoPerComp contains the relative information content of each output channel.
Multichannel input image.
Transformation matrix for the computation of the PCA.
Transformation matrix for the computation of the inverse PCA.
Mean gray value of the channels.
Covariance matrix of the channels.
Information content of the transformed channels.
The operator gen_principal_comp_trans returns the value 2 (H_MSG_TRUE) if the parameters are correct. Otherwise an exception is raised.