damnian wrote:
I don't think FreeCAP is that easy. Has anybody managed to break it already?
Well, based on a similar implementation problem, freecap can be broken.
You can request the same code to be done in a diffrent way by extracting the freecap image URL, and then re-calling the image.
Take, for example, this freecap example image (that I plucked off a site):
We can now try extract just the letters. This is a trivial task, since any background can be seperated rather easally with one pass:
Using a second pass, you can clean the image up enough to get rid of stray pixels:
Next, you can manipulate the image to find points that match the letter. The smaller the red, the less confident the process is that the point it is creating is one that should be used (and can be ignored by the next step);
After that, the points are created in a new image, and linked. Here you can see that a small red dot was ignored because it wasn't confident enough to use it:
Finally, the red dots are removed, and the process turns to attempting to identify the letters. Any letters that it is unsure of will be double tagged (i.e. {first guess}|{second guess}[|{...}]). Only letters which are 85% matched or higher are considered sucessfull and printed:
Trivial
NeoThermic
