yes you can, but its very slow.
you need a canvas to capture the point you're selecting (im pretty sure it can be one pixel in size and still worm for minimal vram usage) then you need to "copy from sprite" with the image manip, then you need to use the getR getG getB expressions which are a part of image manip once the image has been copied. thats pretty much it
the image manip doesnt use screen coordinates when it copies, so 0,0 is not going to be 0,0 on the layout in the getR/G/B expression, its going to be 0,0 on the currently copied image. if your image is larger than 1x1, and isnt placed with its topleft corner at 0,0 on the layout, you going to need to transform coordinates so that theyre "local" to the images topleft coordinates instead of the layout.
just ask if you're having trouble getting it to work because explaining it like this is kinda cryptic.
also note that this series of event is very hammering on the framerate due to its slow gpu to cpu transfer nature so your not gonna be wanting to do it for longer than one frame (another disadvantage to hardware acceleration, albeit not a very big one)