Since 2010 (version 8), Wolfram hasn't made any significant changes in Compile. At this time, vectorized compiled functions and the creation of library functions was introduced. Since shortly after, I was keen to see further improvements and extensions. When I first heard, that Tom Wickham-Jones is working on a "next generation compiler", I was thrilled because it seemed to have an LLVM backend like the Julia language has and it was going to be so much more powerful.

Now, 7 years later still nothing and I'm looking with a bit of sadness at videos like this. Other hot topics, like neural networks, have flourished in a fraction of the time, although everybody who is serious about it probably uses TensorFlow or other alternatives. So it is still not more than a playground for some fancy demos.

@xzczd ↑

@halirutan Hi~

I receive your hint here

Could you give me more information about it?Of course even other language such as c++

@yode Yes, I can give you some information if you like.

I'd like to know that

@R.M. OMG, there is an unknown frog in the chat! How are you?

@yode You need to consider at least 2 or 3 things. One of the major problems is the "similarity metric".

Your examples with thin lines and text contain very few information from an image processing viewpoint. You have to keep in mind: an image registration (or ImageAlign like it is called in Mathematica) uses a similarity measure to decide how it needs to transform the image in order to get a good match.

I think if this two matrix plusdot will get a maxmium after align normally

@yode The problem is not that it gets the maximum at the solution. The problem is that the algorithm needs a good measure all the way until it reaches the solution.

"Your examples with thin lines and text contain very few information from an image processing viewpoint"---In the real case,the empty page will contain more information

You “registration” mean "align"?

I cannot find this meaning in my dictionary..

Yes, "matching two images until they fit onto another" is called "Image Registration"

If you consider using C/C++, then you should definitely check out the ITK

itk.org/Doxygen/html/RegistrationPage.html

It's the library behind many Mathematica image processing functions.

As I know the OpenCV stronger than it?

@halirutan

Or it is better do such align thing?

@yode I don't see anything more in OpenCV than we already have in Mathematica.

Even in the ITK, the only thing more advanced than we have in Mathematica is the "Spline" Transform.

I intend to transfer the opencv function to Mathematica in future,for example it can detect circle..

What I would try out is an elastic image registration, but I don't know if there is anything open-source available.

It is hard indeed as I see.

And it hard to search those English thing for me.If you know that in future,let me know please.

It is a important task for me. :)

@yode Question: The images of the paper with lines that you posted, are these your original images you try to work on?

@halirutan No,there is a sample image.Actually I cannot open it in a open place.But anyway,let it go. :)

@yode My point is: If you have 100 images with those curved lines you need to fix. Then use Photoshop and do it manually. If you have 3000 of those heavily distorted images, then you should think about an algorithm.

Actualy more than 1000.And hard to do that for 100 image in photoshop. :)

github.com/C4IR/FAIR.m

About 10 years ago, they were working in the same field as we were and we met regularly at conferences. His code is for Matlab.

I haven't touched state of the art image registration for a long time, but everything I read on the ITK or OpenCV webpage is known for a long time.

@halirutan Fix that, and this is expected result,and this is task

Thanks I know some Matlab.I will read that.

Good colleague!

@yode If you have the original empty form, you should write an algorithm that finds similar points: Like this here

You get a set of say 30 point to point references. With those points, you could use Spline registration of the ITK

We can use Mathematica to find similar points,but that precision is bad(even worse as I try)

@yode You need to write this on your own.

Of couser,I just want to find some inspiration something. :)

Does anyone get these errors when Mathematica 10.0 (not other versions) starts up?

I think it's some recent paclet update (perhaps issued for a later version) that is causing it.

@Szabolcs Hmm, I can reproduce this as well. Interesting.

I have an odd question: the wolfram language reference online has decided to serve my phone the page in Japanese. Where should I look for a setting to get back to English? I can read Japanese slowly, but skimming for a language setting is difficult, especially when "wolfram language" appears everywhete.

@MarkS. you can add .en to the end of ref page URLs to get the english version. E.g. reference.wolfram.com/language/ref/Print.html.en instead of reference.wolfram.com/language/ref/Print.html.ja.

@b3m2a1 Thanks!