how to robot recognize basic geometric object in window or ubuntu window

[meInvent bbird]

i am doing a robot to automate window itself with python 

i give some basic png diagram such as square, circle, triangle
and hope it recognize all kinds of square like things in window 
such as textbox of notepad etc

how a robot recognize basic geometric object in window or ubuntu window

i use cv2 template matching method 
and then draw black square or rectangle as a template

however, it can recognize the size of template only

and recognize not a square diagram from capture screen

then i try a black ground and white line of square, it recognize a File in menu

i feel that it need to try all kinds of color, and size in order to recognize
similar object in screen capture diagram

how do a robot recognize many color and size in seconds?