Convolution extension - negative values

[darkownt]

Hi:


I'd like the image.convolution JavaScript command to be extended to be able to utilize negative values in the convolution kernal.

I would also request that the convolution process/interface that is introduced at a later date into the PixInsight core also include the capability to utilize negative values.


cheers
Colin

[Juan Conejero]

Hi Colin,

Image.convolve() can be used with any kernel element values. When a kernel filter has negative values, you must specify a kernel weight explicitly. For example:

Code: [Select]

var sharpen = [ 0, -1,  0,
               -1,  5, -1,
                0, -1,  0 ];

var view = ImageWindow.activeWindow.currentView;
with ( view )
{
   beginProcess();
   image.convolve( sharpen, 1.0 ); // explicit unit kernel weight  
   endProcess();
}
You can specify other weights, not necessarily one, depending on your filter and what you want to achieve.

Open an image and run this little script. It applies a classical sharpen filter.

Feel free to play with different kernel filters. Note that the above sharpen filter preserves flux because the sum of its elements is one.

(edited: fixed wrong filter elements)

[darkownt]

Thanks Juan!

I've added that into my script.

BTW Is there any image format that supports negative values produced by PixelMath?  A-B seems to truncate values where B>A...

I'm trying to generate convolution filters with + and - numbers from images with + and - numbers in them.

I have a script which can take an image and use it as the convolution kernal.  To get my negative numbers, in the script I shift the image by a value (usually the negative of the median) to get my convolution filter just before I apply it.  It would be much easier to use pixel math to create a zero centered + and - image, before applying my script without the shift.

cheers
Colin

[Carlos Milovic]

Hi Colin

If you avoid PixelMath, i.e. manually operate the images/pixels (not sure if this works with JavaScrip... I do that all the time with C++/PCL), any floating point format is capable of negative values support. They will be displayed oddly, but, it works.