New Script: 33 Common Convolution Matrices

[Christoph Puetz]

Dear PI Users,

this is my first PJSR script. It features a collection of 33 common convolution matrices, that can be selected by Menu-Entry and
applied to your picture(s).

It can easily be extended and I plan to do so in future, so check for updates regularly, please.
After installation the Script can be called by Menu->Scripts->Common Convolution Matrices.

Hope it may be useful for you.

Kind regards,
Christoph

[Carlos Milovic]

Good job. One possible future addition could be to draw the convolution kernel, or to display the coefficients. Keep the good work comming


PS: Thanks for the citation

[Juan Conejero]

Hi Christoph,

Welcome to PixInsight development, and congratulations on a nice script!

The script works nicely for low-pass filters, but convolution with a high-pass filter requires fixing out-of-range values.

For example, if you convolve with a sharpening filter such as:


 0  -1   0
-1  +6  -1
 0  -1   0


there can be values outside the [0,1] range in the convolved result. This happens because the convolution routine applies a unit filter weight when there are one or more negative filter coefficients.

There are two ways of fixing out-of-range values: truncation and rescaling. Truncation to [0,1] is the standard procedure; it causes some data loss due to ringing, but preserves the overall brightness of the convolved image. Rescaling keeps all the data at the cost of a loss of dynamics, which is a nonstandard procedure (useful in some contexts, but not in the context of general image filtering).

So I think you should add a call to Image.truncate() in line #399 of your script, just after calling Image.convolve():

Code: [Select]

console.flush();

        img.convolve( conv_matrix );
        img.truncate();

vw.endProcess();
Hope this helps.

[Carlos Milovic]

Long live out of range values!!!! Don't opress them!!!!



Jokes apart, for a high pass filter like that one, this truncation or rescalation makes a lot of sence, while for other operators, like the edge detectors (gradients), or laplacians, out of range values have a very meaningful meaning, that should not be changed.

This reminds me that we have to deeply discuss this topic, on future development (and btw, I can't see why complex images cannot be managed by the UI).

[Enzo De Bernardini]

Nice script Christoph, thanks for your work! 

A small suggestion that might be useful for the future: when there is only one value to compare, instead of using many IF, you can use a single SWITCH. For example:

Code: [Select]

switch( index ) {
  case 1:
    conv_matrix = [ 1, 1, 1,
                    1, 1, 1,
                    1, 1, 1 ];
  break;
  case 2:
    conv_matrix = [ 1, 1, 1,
                    1, 2, 1,
                    1, 1, 1 ];
  break;
};
The result is the same, of course.

Best regards,

Enzo.

[Christoph Puetz]

Hi, uploaded Version 1.2 - added img.truncate() as supposed by Juan.
Further extensions coming soon.
Thanks to all !

Christoph

[Niall Saunders]

Keep up the good work Christoph,

And don't be too concerned if you don't see too many 'downloads' - a lot of folk will pick and choose what they need from these kind of code developments. For example, I have not YET downloaded your script, but I AM keeping a close eye on how it develops. Personally, I will 'jump in' as and when I can devote some time to actually 'playing' with your script - but the whole concept is definitely one of interest to me.