Deconvolution?

[Ken31]

OK Guys, flying a kite here.

A couple of years ago I was out and about around sunset, and just happened to spot Mercury with a thin crescent Moon and (IIRC) Mars. I only had my camera with me, so propped myself against a wall and banged off a few HANDHELD ZOOM exposures (digital of course, but JPEG). It amazed me, but amongst the bright sky and the motion blur you can actually pick out a few pixels that are Mercury (and Mars, and the Moon).

As a rainy day project I was thinking of working out how to deconvolve each image and then stack them. The convolution is known, because there are blurry shapes that ought to be small dots, it would be a case of trial-and-error working out a deconvolution matrix for each frame.

However, since I have landed on this forum because of PANSTARRS, perhaps somebody already knows how to do this?

[michael.h.f.wilkinson]

Deconvolution is problematic because it tends to make the noise explode. In this case you only have an estimate of the convolution kernel, obtained from a noisy image. It is possible to deconvolve, and even to do so without full knowledge of the kernel (google "blind deconvolution"), but the results are variable.

[chiltonstar]

OK Guys, flying a kite here.

A couple of years ago I was out and about around sunset, and just happened to spot Mercury with a thin crescent Moon and (IIRC) Mars. I only had my camera with me, so propped myself against a wall and banged off a few HANDHELD ZOOM exposures (digital of course, but JPEG). It amazed me, but amongst the bright sky and the motion blur you can actually pick out a few pixels that are Mercury (and Mars, and the Moon).

As a rainy day project I was thinking of working out how to deconvolve each image and then stack them. The convolution is known, because there are blurry shapes that ought to be small dots, it would be a case of trial-and-error working out a deconvolution matrix for each frame.

However, since I have landed on this forum because of PANSTARRS, perhaps somebody already knows how to do this?

Try the freeware "CCD Sharp". As Michael says, the noise goes rampant so you need low noise images to start with. I've had some success using it to deconvolute webcam images of close double stars - maybe it should work with your images. I've used deconvolution for spectral sharpening (day job) and can't really see why it can't be used more with astro images, even to remove effects of of drive inaccuracy during long exposures of DSOs for example.

Chris

[NickK]

An overview:

Using Pixinsight:

There are many ways todo this - Michael has a uni background but I've seen people on here use both PixInsight and MATLAB too.

As Micheal has said, the noise from the camera has not been convoluted by the same function so either this is removed or ignored.

[Ken31]

Thanks for the info, I will look into it. For a one-off, free is good.

[NickK]

Thanks for the info, I will look into it. For a one-off, free is good.

Trial period PI is another "free" option.