I recently found this link that got me interested in probabilistic programming.
http://naviablog.wordpress.com/2011/06/23/solving-captchas-using-probabilistic-programming/
I am programming with C# for some years now and I'm currently trying to adapt to this paradigm.
My goal is something similiar like in the blogpost. No Captcha but recognition of aritificial low resolution text with applied anti aliasing.
Of course a big part of the program in the blogpost are probabilistic programming libaries of a higher level then planned for Infer.NET.
So input oft the simulation would be a string (Array of discrete variables).
The string has a y value and every letter a x (x2>x1, x3>x2....)
Output of the model would be in a Array2D. Could I use the System.Drawing.Bitmap und System.Drawing.Graphics to output the
result in the Array2D or do I need to write something like font rendering and anti aliasing with Infer.NET abstractions?
After that I think I can let the framework inference back from a picture to the string.
With relations between the letters (markov chain) I probably could make results even better.
So is this feasible and reasonly performant with Infer.NET?
And how would I wire things together. I have some problems imaging where distributions and other things would fit in here.
