PixInsight Forum

PixInsight => General => Topic started by: Harry page on 2008 April 20 13:33:39

Title: Psf
Post by: Harry page on 2008 April 20 13:33:39
Hi all

I am trying to use the decon option and do not know how to
get a external psf

I am using a starlight xpress sxv m25 colour

Any Ideas anyone

Regards Harry Page
Title: Psf
Post by: Juan Conejero on 2008 April 20 14:25:08
Hello Harry,

Welcome to PixInsight Forum.

An external PSF must be a grayscale image, usually a small image, as a cropped star, or a synthetic PSF model that you have generated elsewhere.

To use an external PSF, first load the PSF image. Then launch the Deconvolution interface and click on the External PSF tab of the PSF section. Drop down the combo box (where it says "No view selected" initially) and select the view that corresponds to your PSF. Now you should see the PSF image drawn on the small square image at the right side.

Once you have selected the PSF image, you can continue applying deconvolution in the usual way, that is, just as if you were using a built-in Gaussian or motion blur PSF function.

If you haven't done so already, I recommend you to read two specific tutorials that we have on PixInsight's website:


Take into account that if your PSF image is not well normalized, or if it includes spurious data (asymmetric features or artifacts that don't form part of the true PSF), deconvolution may easily generate artifacts and unreal structures on the processed result. Use external PSFs with care. Unless you have a very good reason to use a PSF image, I recommend you to stick with built-in PSF functions.

Hope this helps. Let us know how it goes.
Title: Psf
Post by: Carlos Milovic on 2008 April 20 14:26:04
Hi Harry

The most "obvious" source to make an external PSF is to use a stellar image. Build a preview around a star you choose, with the following requisites: as small as possible, to avoid other objects to be included; center the star as much as you can (if not centered, your deconvolved image may shift or show some artifacts); and, use linear data. Now, use the preview to create a new image. Depending on the deconvolution process you use, then there are more restrictions to the image that may be used as external PSF. Deconvolution (Richardson Lucy and Van Cittern algorithms) allows only square, grayscale, images as external PSFs. WienerDeconvolution, by the other hand, may use non square images (and I'm not sure if it allowed color images too.. can't remember right now).

So, in short words, external PSFs are independent images, relatively small, that are used for the deconvolution algorithm. You have to select them from the image view list, displayed at the proper tab of the process window (and don't change the tab).

Other ways to create external PSFs is through a script, so you may "draw" a PSF from an equation or input some fixed values. The same may be achieved with a global PixelMath instance. I'll look for examples of both procedures and post them here.

Finally, we'll work in the next months in a PSF manager for the Convolution process, that most likely will include an option to generate new images, that may be used in the Deconvolution processes as external PSFs too.
Title: Psf
Post by: Harry page on 2008 April 21 10:05:28
Hello gents

Thanks for your reply and I will have a go at this, but could you tell me why
colour psf are not allowed / desirable.
as I use a one shot colour camera Can I use the luminance for the pcf  or would I have to create a
grey scale image from the rgb image I have!
I have also understood that it is best to use a external psf, why does your opinion differ from this?

Regards Harry Page
Title: Psf
Post by: Juan Conejero on 2008 April 22 10:23:38
Hi Harry,

why colour psf are not allowed / desirable.

as I use a one shot colour camera Can I use the luminance for the pcf or would I have to create a grey scale image from the rgb image I have

Color PSFs are not supported in our present implementation, mainly due to technical limitations.

However, unless you have some exotic optical problems in your imaging train, it is very unlikely that you ever need a different PSF for each individual RGB channel of a one-shot color image.

Of course, you can extract your PSF from the luminance of your original image (e.g. from a small area that includes a star). This will be a grayscale external PSF that you can use to deconvolve your RGB image: the same PSF will be used to deconvolve each channel. To do this, you have to uncheck the "Luminance" check box on the Algorithm section of Deconvolution. However, doing this is usually a bad idea.

Deconvolving the three RGB channels separately is generally an error. Usually it is much better to deconvolve just the luminance, leaving the chrominance intact. Due to the characteristics and limitations of the human vision system, the luminance is responsible for almost all of the detail perception. For this reason, deconvolving the luminance and chrominance together will increase noise (by transferring chrominance noise to the luminance) and provide no additional detail improvement.

Deconvolution must be applied to linear images, that is, before any nonlinear transformation, as a histogram stretch. Keep in mind that deconvolution should be applied with a physical justification. No PSF can be valid for all pixels of a nonlinear image simultaneously. To deconvolve the luminance of a linear, one-shot color image, you must check both the "Luminance" and "Linear" check boxes of the Deconvolution interface.

I recommend you to read this tutorial, which includes a practical deconvolution example:


The example above is with three separate narrowband images, but it provides a lot of important information and shows you many techniques that you can directly apply to one-shot color images. The main difference is the "Linear" option of Deconvolution that must be enabled in your case, as I've said above.

I have also understood that it is best to use a external psf, why does your opinion differ from this?

Because it is very difficult to extract a good PSF from the image itself. In most cases, it is almost impossible. This is because any subimage extracted from the original data will generally be affected by local irregularities, as noise and other spurious structures, that will invalidate the extracted PSF unless it is further transformed.

It is much more efficient and accurate to use a synthetic PSF based on a model built from direct measurements. For example, if you know the FWHM for a set of stars in your linear data, then you can use the FWHM value to derive a synthetic Gaussian PSF. This is the standard procedure for deep-sky images.
Title: Psf
Post by: Harry page on 2008 April 22 12:41:28
Thanks for the insight into my psf question and will take all advice in hand and try to implement
the suggestions

Many thanks for the help

Regards Harry