Gaussian Blur

[roberthurt]

I'm relatively new to Pixinsight but am starting to use it on our public affairs imagery for NASA's Spitzer Space Telescope in conjunction with Photoshop/FITS Liberator plugin.

One tool that I use regularly that seems to be missing from the arsenal is a simple image convolution function. Gaussians are of particular utility, as often working with infrared data it is useful to degrade the resolution of higher-res images to better match composite overlays. There are real advantages to doing this before stretching the image (which is generally not feasible in Photoshop, but would be perfect in PixInsight).

Unless there is a way to do this that I've missed, consider this my wish list item! 

[Nocturnal]

Hi Robert,

welcome to PixInsight! I've never had the luxury problem of wanting to soften my image so I haven't looked for this feature yet There are two processes under 'Convolution', could these do what you need? I'm sure Juan or someone else with more knowledge of this subject than me will answer soon.

[georg.viehoever]

If your goal is to "soften" the image, using the ATrousWaveletTransform with layers 1+2 or so disabled is also an option.

Georg

[Harry page]

Now Thats just showing off

Regards Harry

[Juan Conejero]

Hi Robert,

Welcome to PixInsight Forum!

Indeed we lack a basic deconvolution tool. I plan on writing one after Summer. It will be similar to the current MorphologicalTransformation tool, i.e. it will include a graphical kernel filter designer interface.

As Georg has pointed out, a wavelet transform can be used to implement an efficient low-pass filter, and is much more flexible and controllable than a convolution.

The tool of choice is ATrousWaveletTransform. Just disable several small-scale wavelet layers, as required. If you need more control, you can vary the scaling function used to perform the wavelet transform. Peaked functions (3x3 linear interpolation, 3x3 Gaussian) are better at isolating small-scale structures, so they can be used to control the achieved smoothing effect. Smooth functions (B3 Spline for example) can be used to apply more aggressive low-pass filtering.

An alternative is a script to implement a convolution, either in the spatial domain or in the Fourier domain. This is very easy to do with JavaScript in PixInsight. I'll try to write a simple script tomorrow, which can be an excellent programming example by the way.

[Jack Harvey]

I have softened some of my images by setting the first and second tier of Atrous Wavelets ot a negative number like -0.1 or -0.2.  Do not know the math but this appears to give a gaussian type blur to my images.  Have to admit I rarely use it, sometimes with a star mask to soften the stars.

[vicent_peris]

Hi Robert,

instead of convolving your data... if you have high SNR you can try to deconvolve the lower resolution images.



Regards,
Vicent.

[Carlos Milovic]

See at the announcements section. GaussianBlur has been released as a new module. It is a very simple implementation, but useful.

[darkownt]

I'd like the "basic" convolution tool to include (in addition to the core kernal editor/matrix dialog etc.) the same tabs as the deconvolution/restoration tools.

1. Gaussian Convolution (size and shape settings etc.)
2. Motion simulation convolution... (linear)
3. External file (greyscale or color image? used to convolve the target)

I'm particularly interested in the external file capability since it allows for experimentation with various measured and simulated PSFs and thier effects on simulated or true images, and could help greatly for learning how to tune deconvolution using the same PSFs.  etc.

This also opens up the possibility of non-linear (arbitrary path) motion convolution/deconvolution testing.  Not that it matters for long exposure astronomical pics which are done right but... interesting and possibly useful.

On a side note it would really be cool to use "star test" simulated images generated from aberrator (http://aberrator.astronomy.net) on "true" images (hubble etc.), scale them down to sub million dollar astronomy magnifications and see what can be done using deconvolution and what settings are best for what kinds of aberrations (inside focus vs outside focus etc)

cheers

Colin

[Carlos Milovic]

I think that the Convolution tool will be less complex in the interphase as the other tools, so you'll have a process for each task. Gaussian blur has been already done (1.6 release will include the oficial module). Motion blur may be the next addition along with the external file one... Count with them for the next release too. I'll work again on code development this december.

[RBA]

Gaussian blur has been already done (1.6 release will include the oficial module).

It didn't, AFAIK.     

[Carlos Milovic]

As you know, there were several things that had been left in the 1.6.0 release. Some of those finished modules, and a pair of two new ones (I already wrote and send Juan one) are supposed to be included in the 1.6.1 release. If not, prepare the flames for the riot.