@Superku:
I have read the article. So for each pixel, the vector is calculated an the required pixel(based on depthmap) into the calculated direction is copied directly from the screen image?

So when you have a box floating above the water, you can not look under it by looking at the water? So minor perspective erros will occur?