Well, the simplest approach is to simply take each line of pixels and scale it up to the same width as the widest row. That gives you a perfect rectangle. If your aim is to recreate then image on the 'plate,' then you'll need to apply some vertical stretch as well.

Well, if you're willing to create an image that is smaller than the widest row, then you can use a 'shading' approach.

You specify your desired resulting image width/height in pixels. Then, loop over each pixel, and 'sample' the original trapezeoid.

You'll need a function to convert perfect-rectangle coordinates to trapeseoid-coordinates; a simple interpolation approach will give you reasonable results, but you may want to build some kind of perspective correction into it.

The simple interpolation approach is to take the original coordinates as fractions of the target image dimensions (e.g. pixel (20, 20) in a 400x400 image gives you (0.05, 0.05)). Take the four vertices of the trapezeoid in the source image and use these coordinates to interpolate between them.

If you need really high performance, you might be able to adapt this technique to the GPU, though if the images are too large then you might run into problems.