Efficient Bot (AI) Theory

[MichaelSpectre]

After researching how some of the bots here work, I started implementing my own.
I'm using C#/C++ with DLL injection, allowing for in-game overlays.
Works fine so far. I'm able to launch/inject the process, draw an FPS counter, and capture the backbuffer.

But as far as efficient AI algorithms go, I'm a bit lost.
Are these bots using some sort of image comparison techniques in order to know what to do next?
(i.e., where to move the mouse to, where to click, which keystroke to simulate, etc...)
I mainly question SarkothBot, in which a set of uniformly-scaled images are required for the bot to function properly.
I simply fail to see any algorithm where a 10x11 image of the letter 'a' can be compared against the screen, without a major downpour on the framerate.

If I were to implement a similar approach, where should I place my research?
Alternatively, what other ways could I use the backbuffer to aid in my bot's AI?
Many thanks to any who can point me in the right direction.

[cha0sbg]

The sarkoth farm script uses just places to click and check for pixel color to know if it should continue or not.
As for the imagesearch if ur not planning to inject anything in diablo it's fine, but if you plan to inject ur dll as you said you should go
all the way.

[MichaelSpectre]

So I guess in theory, I could use the backbuffer dimensions to determine to current screen settings, and have a respective list of points for each resolution lined up?
I also have a nice image comparison library that could probably do the job a bit quicker than Sarkoth, as long as I'm not parsing the entire backbuffer for a single partial image.
Perhaps just have a list of rectangular points to check for the image?

As time progresses I'll probably be able to grab data from the pixel shaders to gather more accurate information.
I ran Diablo III in Pix for Windows and found an abundance of decent resources that could, at the very least, determine your current class.
But for now I'm just looking for something simple and efficient that actually works.

As far as DLL injection goes, how detectable is it? I'm using EasyHook (instead of Detours, or the typical method of replacing the Direct3D dlls with custom ones).
Also, is there anyone else on these forums using similar methods? Or are they mostly scripting via a 3rd party app?

EDIT:
I should probably elaborate why I'm using DLL injection...
Most of the scripts I've come across don't allow for true fullscreen modes, or only support a small list of screen resolutions.
Some scripts just have too much overhead involved, making the game a bit choppy.
Also, I enjoy the idea of having an overlay, providing the user with some eye-candy. =)