The other two examples can be seen at
http://www.geocities.com/jgroveuk/voynich/ColSep2.jpg
http://www.geocities.com/jgroveuk/voynich/ColSep3.jpg
The algorithm deals with background colours by assuming that the colours which were picked from the image are effectively a combination of two colours - in our case the raw vellum colour and the pigment. The vellum colour is separately picked and marked as background, and the algorithm then subtracts that colour from the other basis colours before analysing the image.
So the image might be split into components that reflect the vellum colour, pure green pigment (rather than green+vellum colour), and pure brown ink. Removing the green pigment from the final image will then show a better image of the underlying colours.
The issue is made a bit more complicated with the VMS because the vellum colour is a rather similar hue to the brown ink, which makes it difficult to use both in the analysis. Often the result is better if you choose just two colours (e.g. green paint and brown ink) and let the system generate the third orthogonal colour. Typically there will be very little signal in the generated colour, but including it in the final output can make a visible difference to the image. There is a certain amount of trial and error in picking appropriate colours, but if you find a good set the results can be surprisingly good.
Jon.
> -----Original Message-----
> From: Gabriel Landini [mailto:G.Landini@xxxxxxxxxx]
> Sent: 06 July 2004 16:46
> To: vms-list@xxxxxxxxxxx
> Subject: Re: VMs: Colour separation
> Importance: Low
>
>
> On Tuesday 06 July 2004 15:06, Jon Grove wrote:
> > For those interested in uncovering the overpainted text
> with an algorithm
> > similar to the colour deconvolution used by Gabriel
> Landini, I've created a
> > plugin for PaintShopPro (and Photoshop), which will perform colour
> > separation. The results on the image used by Gabriel can be
> seen here:
>
> Very nice too.
>
> > The colour separator effectively analyses the image into a
> new colour space
> > using colours specified by the user. If insufficient
> colours are specified,
> > it will create some 'orthogonal' ones to aid the analysis.
>
> Yes that is more or less what Ruifrok's method does.
> Can you post the blue image as well? (I do not have Photoshop
> or Paintshop and
> I think that the Gimp cannot run plugins under linux).
>
> Cheers,
>
> Gabriel
> ______________________________________________________________________
> To unsubscribe, send mail to majordomo@xxxxxxxxxxx with a body saying:
> unsubscribe vms-list
>