Software Bisque

Take a virtual tour of the Software Bisque campus Watch Steve Bisque discuss the Paramount mount with S&T's Dennis Di Cicco Read the Paramount MYT Review Read the Paramount MX Review

Hot Pixels 2: The Heuristics
Richard S. Wright Jr.'s Blog

I know it sounds like the name of a bad Sci Fi sequel, but as of daily build 10229, the Starlight Xpress plug-in has a new hot pixel elimination feature. The old hot pixel feature is still there, but I'm encouraging users to try out the new "Heuristic" checkbox instead. Please do, and report back to me on this blog thread (not on the forum buried in the middle of a discussion about STX-AO-USB4/Ethernet-Tachyon problems... because believe it or not I don't read all those posts ;-)  If this works as well as I think it will for other users, we'll see about rolling this up out of the plug-in and into TheSkyX itself for all guide cameras, X2 and ASCOM. This is ESPECIALLY intended for guide cameras by the way. Let me explain...

So, this is a blog, I can be personal right? Well, I'm almost embarrassed by how simplistically brute force the first approach I came up with was. It also really put too much burden on the end users to figure out what an appropriate threshold should be, and it varied from camera to camera. It did work, but just because something works doesn't mean it's a finished product. It was also confusing to some customers, and I all but had to argue in one forum thread with a user who I could not convince that he was destroying his data by being over aggressive with the threshold value. I'm a software engineer, and graphics and imaging have been one of my specialties since before I got into this industry... so I hope some people will give me the benefit of the doubt when I tell them something that might seem counter-intuitive.

This feature is designed to solve two problems, and two problems only. One is calibrating a guider that has no shutter and thus no way to eliminating bright pixels that may be mistaken for a star during calibration (the dreaded "star did not move" when looking at the image, the star obviously DID move!), and secondly when guiding of course, we want to find the center of a nice guide star, not get stuck on a hot pixel mistaken for the center of a bright star. It is even possible to have hot pixels inside a guide star! The previous method would notably fail on this rare, but not impossible scenario.

It is EXACTLY a hot pixel removal tool. It IS NOT:

1. A warm pixel removal tool.
2. A "sparkle" removal tool.
3. Auto-magic cosmetic correction.

 With this turned on, you will still see many "warm" pixels, those bright white blocks on the screen stretch off to the side of a guide star or in the frame while looking for a guide star. A survey of different hot pixel removal techniques all had one shortcoming, and that was they were obsessing too much over detecting any pixel that looked out of place. The fact is, we only need to find pixels that can be mistaken for a star, not all of them. The algorithm (and no, I'm not going to spell it out for our competition<vbg>) is a little more CPU work than the brute force method, and looks at localized variations in the background. It's still very fast though, even on the Raspberry Pi builds. No user input required at all. Stars also have a "signature" if you will that is different than a lone hot pixel. The new technique only removes pixels that can be mistaken for stars, and it even will find a hot pixel inside the radius of a bright star (well, it should).

I've put this in the Lodestar plug-in, because it is one of the most popular cameras we support without a shutter, and hot pixels from this camera are a frequently encountered issue. The previous 'fix" all but eliminated the problem, but I think the new approach will do even better, and will require less engineering effort on the part of the end user. The goal is, it should "just work".

The test is simply to do some guider calibrations with only this check-box on. If you can calibrate repeatedly without getting stuck on a hot pixel then you win the prize (the prize is trouble free calibration). I've calibrated, and recalibrated all over the sky with my Lodestar, which was tricky without darks or the old hot pixel technique before, and as far as I can tell the new method works great. Time to let some guinea pigs... um... beta testers give it a try! If it fails, if you could turn on "Save images" for the autoguider and send me some images where the hot pixels were a problem, that would be very helpful too!  Normal guiding should of course be unaffected. Also, don't be mislead by a bright white pixel, our screen stretch enhanced contrast a lot (something else I'm tinkering with in the labs). The better indicator of a truly hot pixel is the 3D graph on the guider graphs. They will show up as an isolated spike, and if I did my job right you should never see those again.

Remember too, when you select a guide star, pick one that is not crowded with several other stars nearby. There is no "nearby bright star" filter... yet.



Posted 09-06-2016 2:41 PM by Richard Wright


© 2017 Software Bisque, Inc. 862 Brickyard Circle, Golden, CO 80403-8058 USA - phone: +1 303 278 4478 fax: +1 303 278 0045