Identifying something on the screen

I'm curious as to how the CodedUI Test functionality in Visual Studio works.

I can see how it can identify keypresses and what's typed by querying the keyboard state, but I'm having a hard time figuring out how (i.e. code-wise) it identifies what button was clicked, where that button is located, and how to track it if between the recording and test runs the button has mov开发者_高级运维ed or happens to be scrolled off the screen. I'm mainly interested in items which are foreign to me (i.e. I don't have access to the source code and/or it's running on a different process from my programs process).