Deconvolution Processing

Deconvolving an image can result in better sharpness and detail. This procedure works best with well-sampled or over-sampled images. In other words, images that have a small pixel scale. For example, in the Whirlpool Galaxy image used here, the pixel scale is 0.73 arcseconds/pixel. With typical 2-arcsecond seeing conditions, this means each star is about 3 pixels across. This is a well-sampled image and can be deconvolved with success. Images taken with a short focal length telescope such as a refractor or HyperStar SCT may not yield ideal results.

Deconvolution works by determining the Point-Spread Function (PSF) of a star in the image. This is effectively the shape and size of the star. Deconvolution tries to improve the PSF of the star, tightening the star image and resulting in more detail. Below is the profile of a typical star, showing what is called a Gaussian profile shape.

Below is the initial image of the Whirlpool Galaxy. It has been DDP stretched, as shown on the previous page.

Go to Filter > Deconvolve to open the Deconvolution window. The window starts in the Noise Model tab. Click on Auto-Extract under Noise Extraction Tools. This will automatically determine the background noise level.

Select the PSF Model tab. Choose Gaussian under Function Type, then press the Select From Image button. Mouse over the image and click on a non-saturated star. The Information window will be open and will display the Maximum value of the star the cursor is over. Make sure this is less than 65000.

Inevitably the PSF value chosen will be too large. If you run the deconvolution now, you will get an overprocessed image, typically with black halos around the brighter stars. This is shown below.

To avoid this, after selecting the PSF from the image, reduce the value. You may have to try a few iterations to see what works best, but usually a 50% reduction is a good starting point. Here the value is reduced to 2.2.

To run the deconvolution, select the Lucy-Richardson method in the Deconvolve tab. If processing a large image, you probably want to start by processing a subframe around the interesting part of the image to test the result before running the full image, which can take a very long time. Under Operate On, select a width and height for the subframe (512 x 512 is usually fine) and drag the subframe box to the desired area of the image. The number of iterations is a matter of trial and error, but 6 to 10 is a good start.

Below is the result of a 10-iteration deconvolution with a PSF radius of 2.2.

Return to MaxIm DL Page