As far as I know K. Bouman [0] was the scientist leading the charge on the image processing/reconstruction. A few of her later papers probably have hints [1, 2] about how this is done, but I haven't seen the official release.

[0] http://people.csail.mit.edu/klbouman/

[1] https://arxiv.org/pdf/1903.08832.pdf

[2] https://arxiv.org/pdf/1702.07361.pdf

Thanks, I think I added the link to the paper while you were writing this. From skimming it I can't tell the answer to my question though.

How do they know what appears "black" in the image is really black vs. relatively black?

If you observe a very wide band of light (e.g., EM radiation) and there is nothing received from those spots, then, for all intents and purposes, the region of that image is black.

Now, if you're asking if, perhaps, the region isn't really black, but rather it's emitting some sort of small radiation relative to the bright region, it would be essentially impossible to know without much higher resolving powers (since it might even be indistinguishable from the background noise generated by the surrounding region). There is no way to really know if it's "perfectly black" vs. "orders of magnitude darker than the surrounding regions."

Makes sense. So I guess they could probably tell us an upper bound on how bright it is.

Indeed! That’s for sure: it’s probably not hard to extract a bound on the magnitude of a part of the spectrum from this analysis.

Any idea how bright it might be? Eg, could it be as bright as the sun? The moon?

The project's director was Shep Doeleman

https://news.harvard.edu/gazette/story/2019/04/harvard-scien...

Indeed, from what I gather from this thread and external links this “image” seems to be more of a plot than an image. In fact, are we looking at a matplotlib plot of the data with a “hot” colormap? The iopscience paper even references matplotlib. I’m just making some educated guesses here, but it’s still fun to think about. BTW: Python for the win! iopscience paper makes several references to Python tools (e.g., numpy, Jupyter etc.)

Every image is a plot... This one just has had a bit more processing gone into it than your average demosaiced and noise reduced vacation photo.

Another interesting thing about that iopscience paper is that you can turn MathJax on/off with a link at the top. Mathjax is off by default, but when you turn it on, you get a nicely rendered equation instead of a gif image.

Summary of their summary:

“arising from the rapid atmospheric phase fluctuations, wide recording bandwidth, and highly heterogeneous array”

(Filtering out bad data)