How to model mapping between two images? (Python, OpenCV)

Question

I want to create realistic like license plates, I write a code to create them but they are not very realistic, for this I want to model mapping between real license plates and fake ones and then map fake license plates to convert them to realistic images.

My fake license plate:

And this is its correspondence in real world:

I want to find a suitable mapping that convert fake LPs to real one, but as you can see the mapping is not just spatial (I cannot model it just by a homography transformation). How could I model this kind of mapping?

Answer 1

Perfect task for GAN networks. Check this project for example CycleGAN .