|
You could make an artificial star image and find the best parameters by trial and error. Here is how, assuming the convolution kernel is reasonably well approximated by a Gaussian (it usually is):
- Start with a small blank image, all black. Make one pixel near the centre of the image white.
- Gaussian blur the image with a radius of your choosing. There will be one radius that works best. This is the guess, trial and error part. It's pretty safe bet that the optimal value will be between 1 and 10.
- Normalize (contrast stretch) the brightness in the image (might not be needed but at least it will help you better see what's going on) - make sure you do not over-stretch, ie clip the brightest or darkest pixels.
- Now you have your "star" image to feed the deconvolution routine (you can cut-n-paste into your other image or do whatever works with your processing software)
- keep trying again with different blur radii till you find what works best.
|