Condensation Algorithm - Tips to script this one?

[darkownt]

Hi all:

While playing around I came up with a condensation algorithm to reduce (condense) the size of stars.  It uses an intensity conserving transfer function weighted by a product of pixel brightness (and kind of looks like the gravitation equation m1*m2/r^2).

I have attached a pixelmath process (NOTE:  rename the image to "startest" to use it) which performs the algorithm using an inverse square law and over a 5x5 pixel region.  I also have attached a document showing the formula I want to use to generalize this into a script.

Also, can I somehow remove the reference to the specific image, renaming images is a pain...

Let me know what you think or if you think it's useful.

PS:  The key usefulness of this algorithm is revealed by multiple applications/iterations.

Cheers
Colin

 

[Nocturnal]

Colin,

use $T in PM to indicate the target.

 

[Carlos Milovic]

Hi Colin

Just at first glance... I don't think that this is too away from a deconvolution (or, a high pass convolution). You may try to generate a sinthetic image with a radial profile following that R^-2 decay, and use that image (normalize it so the total value is 1, Average*number_of_pixels).

As I said, this may also be done as a highpass filter. Since you are analysing a neighbourhood, you may calculare the gradient at a givent point, and create a function that rescales the image acording to that gradient. Other alternative may be to work directly on the gradient domain, and then use the Poisson solver Georg suggested to get the image. If you follow an iterative approach, you may emulate the 1/R^2 model with some success.