Wednesday 20 April 2016

Pattern Recognition

Technology marches on. Sometimes I find myself caught entirely unawares by some capability which not only works flawlessly, but which has become almost freely available to the consumer or enthusiastic amateur. I still remember buying a digital SLR camera which had face recognition built in almost as an afterthought. I'd given it no great thought until - with the camera sitting powered-up on a coffee table - I noticed that it was locking in on the picture of the queen's head on a five pound note! It was a shock to realise that this supposedly futuristic image-processing functionality was not only highly robust, but so cheap to implement that it was barely mentioned in the camera's sales material.

I had a similar experience earlier today. After a long run of cloudy nights, I've finally been able to get outdoors with the telescope and attempt to continue my long-running adventure in astronomical imaging.

This being Spring, one of the obvious candidates is M13, the globular cluster in Hercules. It's easily visible in telescopes and not hard to find in binoculars. I've looked at it many times over the years, but only this week did I manage to get a picture of it.

I shot this using a telescope on a GoTo mount. The mount contains a computerised database, and provided it's set up reasonably well at the start of a night, the "kit" enables the telescope to automatically locate and track astronomical objects. So, although I can find M13 for myself, I didn't need to: I just typed "M13" into the handset and off it whirred. The pointing accuracy was such that the cluster ended up close to the field of view and there was no difficulty obtaining images,

M13's fairly easy game, though, and I fancied a stiffer challenge. High in the Spring sky, near the "handle" of the Plough, is M101 - a "grand design" spiral galaxy of quite exceptional beauty. It's also very faint and a reportedly tricky object to see by eye alone, even through a telescope. With light pollution and a bright moon, there was no chance of that - but I was still optimistic that I could obtain an image. After all, at least I didn't have to worry about the pointing part - the telescope and its GoTo mount could take care of that for me.

However, I didn't succeed. Here's one of several frames I took over two clear nights:

Fascinating stuff, eh. Some random stars and electrical noise, and no sign of anything resembling a spiral galaxy. Now, I mentioned that the GoTo mount does need to be set up properly at the start of the night - accurate polar alignment and all that - so there's always a possibility that it isn't quite aiming in the right direction by the time it thinks it's found M101. Compounding that, the digital camera that I attach to the telescope samples a smaller field of view, so the object of interest might be falling just outside the camera's reach.

It was then that an astronomer on Twitter, Andrew Gray, mentioned that there is such a thing as "plate solving", The idea is that a sufficiently powerful pattern matching algorithm can analyze an image of some stars and work out exactly where in the sky they correspond to. That's an incredible feat of computation, especially if you don't know in advance little niceties like angular scale, brightness calibration and so on. Now, I was distantly aware that something like this capability existed, but I had no idea it was now easily within reach of the amateur such as myself.

However, within minutes I had uploaded my frame to:

And almost as quickly I had a set of annotations for my frame:

This is ASTONISHING. Not simply because it comfirms that M101 was indeed within my field of view - but presumably too washed out by glare to show up - but that this capability exists and is free for use, and returns results quickly enough to be useful for astronomers actually sitting at their telescopes, trying to find stuff. I didn't even have to submit the image in some fancy, astronomy-only image format, either - I just sent a Jpeg.

For the sheer hell of it, I also ran my M13 image through the processor:

I'm amazed, and impressed, and excited by the possibilities. Truly we are blessed to be living at a time when such miraculous feats of technology are easily within our grasp.

Just to end, here's a close-up of M13:

The light I caught had travelled 25,000 years to reach my telescope. If there's ever a day when that sort of thing doesn't send a shiver down my spine, please feel free to shoot me.


  1. Well, "plate solving" goes back at least a century. (I guess you never had to work out plate solutions as a student? I actually did that using micrometer measurements on an old photographic plate in an undergraduate astronomy class ;-.)

    What amazing about, of course, is that you don't have to give it the correct coordinates of the individual stars in your image; it combines built-in star catalogs with the really impressive library of compressed star patterns so that it can figure out which stars are in your image (though it's faster if you give it some starting guesses).

    (I know that at one point the folks were testing it by grabbing random night-sky photographs off Flickr and seeing whether the code could get a solution...)

  2. Very cool! Do you have to travel away from your home to take these images, or do you live sufficiently far from city lights to allow for true backyard astronomy?

  3. All these images were taken on the decking outside my garden office. If I use a USB extender, I can even be sitting at my PC writing while acquiring images on a laptop plugged into the telescope.

    But it's not a particularly good site. There are bright street lights nearby, and a sports field which is often lit up, as well as general light pollution and poor cloud cover. The advantages of being able to set up my scope and start imaging in minutes, rather than having to pack it all into a car and drive somewhere, are more than adequate compensation.

  4. I entertain guests and myself using an augmented reality app star app for my phone. I'm not good at memorizing constellations so it's a big help. I can usually spot Mars, Jupiter as well as the obvious such as Orion, Cassiopeia and Ursa Major. The app has helped me learn several more. It's a lot of fun when I'm in the mountains where there's low light pollution and you can actually trace the Milky Way across the sky.

    My girlfriend gave me a telescope for XMAS and it has an adapter for the phone but it was incredibly hard to assemble and get it aligned. It's always a bit out of focus. Good to know this pattern mapping service exists should I ever be able to snap a shot of anything in focus :)