I've been working on developing a new image stacking plugin for MaximDL, based on the Drizzle algorithm pioneered by the NASA team for Hubble.
Matt Thomas was kind enough to make some sample images available for me, so I've been testing with good quality source frames. Using these frames, I managed a 1 pixel reduction in FWHM, which corresponds to a reduction of about 2arcsec at the final image scale of 3200x2400. I've also been in communication with Bob Denny, the author of PinPoint, who graciously made available to me knowledge of the inner workings of the PinPoint image distortion model. With this knowledge, I've been able to incorporate image distortion compensation into my Drizzle plugin, which was obviously one of the original design goals of the algorithm when NASA came up with it for working with the early distorted Hubble images prior to installing the corrective optics.
I've put a copy of the Drizzle plugin up here for beta testing, although it
does have some limitations at the moment.
As it turned out, getting the Drizzle stacking working was pretty easy, as was the image distortion compensation. What has taken me many nights of work and I'm still not sure is working correctly is rotation compensation. Thus, in this beta version, you can perform image rotation and distortion compensation, but I'm not sure how well both together works. Also, image scale compensation isn't tested yet, so all images must have the same plate scale currently.
Please note that the result window is created immediately the plugin starts
since it's used to build up the result image into, but the plugin will display
a message box once it's finished processing.
Depending on your system spec, image size/count and parameters selected it could take several hours to run.
The Drizzle stacking is a very computationally expensive process and I must admit I've not spent any time on optimisation yet, I've been focusing on getting it working!
You may wish to select and process a small subframe for your first run, just to get to see it working.
The images to stack must all be plate solved by Pinpoint (or PinpointLE). The Pinpoint solution is used for translation, rotation, scale and distortion compensation.
Due to the way the Drizzle algorithm works, the input images must have subpixel dithering, so autoguided images must have dithering enabled.
Using higher order plate solutions will result in better distortion compensation. This is selectable in the Pinpoint advanced settings.
All images to stack must be in memory - the plugin will operate on all currently loaded images, so only load those you wish to stack.
Colour images cannot currently be Drizzle stacked, so you will have to split them into RGB and stack them individually then recombine.
The currently selected image when the plugin is run will be treated as the reference image with all others being aligned to it.
If a subframe is selected on the reference image, processing can be restricted to it, to speed up testing different drizzle parameter settings.
Once processing is complete, the output image labeled "Drizzle" is displayed, along with another image showing the weight distribution of input pixels over the output pixel grid. For best results input parameters should be selected to ensure a reasonably even weight distribution.
· Drop Size = the fractional size of the input pixel used to calculate the output values. This should generally be slightly larger than the Output pixel size.
· Droplet size = the fractional size of the drop used to determine the overlap onto the output pixel grid.
· Output Pixel size = the fractional size of the output pixel compared to the input pixel. Using a value of 0.5 will result in the output image dimensions being double those of the input image. Selecting a smaller output pixel size generally requires more input frames to ensure a good coverage of all pixels.
· Use subframe = only process the selected region for speed of testing different settings.
· Correct Distortion = Use Pinpoint distortion map to correct input image distortion.
· Correct Rotation = Rotate all input images to have the same rotation as the reference image.
· Correct Scaling = Rescale all input images to have the same plate scale as the reference image.
v1.0 - Initial release: rotation not well tested with distortion, scaling not tested.
v1.2 – Add dec compensation to the alignment algorithm. Still some alignment problems in some cases.
If you download this to give it a go, please post feedback on the Yahoo
group, or email me directly at: winfij_AT_gmail_DOT_com
One final note - if you have any errors, please note the exact error messages and let me know!