It depends. If there are only a few drawings and they will probably never change, then using images will be pretty simple to implement.
But, take the project I’m working on right now: I have 100+ conveyors to display, all of which are individually controlled. Some are straight, some curved, some are belt style and some are link style. They cross each other, sit on top of each other, and are obscured by other devices. So, by using the paint component, I just draw the belts using parameters that I pass into a function (length of the belt, radius, etc) and it becomes a bufferedImage that I draw on the canvas. And by using things like transparency, I can make belts semi-transparent, completely invisible, or give them tints and shading to indicate status. And, since I treat them as objects, they can become buttons, labels or whatever. And in huge plants, I can zoom in and out and pan, like Google Maps. The possibilities are endless.
The disadvantage? It was pretty freakin’ hard to figure out in the beginning. If I wasn’t going to be doing a lot of projects and didn’t need this flexibility, I would have just used images.
At the end of the day though, I don’t care how anyone does it. FPMI blows away any other software for screen design.