Principal Component Analysis

[markgritter]

A similar technique to NNMF from my last entry is Principal Component Analysis. It still factors the input matrix, but now negative values are allowed. A big challenge for using PCA is what to do with those negative numbers, which have no standard interpretation. (This is one of the advantages of NNMF, that it can be applied in scenarios in

Comments

[markgritter]

Trying it on a larger set produces results that are... not boring, at least?



I kind of like this jalapeno flower, but it is still "muddy"

[markgritter]

It might be better to do PCA on colors individually, or red/green/blue/alpha color channels individually.