[ptx] Subpixel accurate template matching algorithm ???

[alexandre jenny]

> This sounds like the problem I attempted when I wanted to 
> write a lens 
> distortion specifier:

That's exactly what I want to do.
 
> I printed a big (A0) sheet containing a rectangular grid, 
> then tried to 
> automatically locate the coordinates of all the grid junctions.
> 
> I couldn't get this to run solidly enough, so instead I did it mostly 
> manually. :-(

I really want this system to be automatic.
 
> PS. The best method I did find was to use a small template cross, 
> convert the input image into B&W, and then do a convolution 
> between the 
> template and target image. Even without any rotation of the template, 
> this did tend to find useful maxima close to the proper 
> crossing points.
> 
> An algorithm that knew how the target image was supposed to 
> look could 
> find each horizontal line, scan along it, an d determine where the 
> vertical crossings were.

Yes in fact, I was thinking of both algorithms but couldn't decide.

My goal is to create tools to achieve something like in DxO labs is doing : 
http://www.dolabs.com/Photography/DXO_Optics_PRO.htm

Alexandre