Basic Knowledge

affine transformations

https://kevinzakka.github.io/2017/01/10/stn-part1/

When an image undergoes an affine transformation such as a rotation or scaling, the pixels in the image get moved around. This can be especially problematic when a pixel location in the output does not map directly to one in the input image.

To solve this problem:

In the illustration below, you can clearly see that the rotation places some points at locations that are not centered in the squares. This means that they would not have a corresponding pixel value in the original image.

So for example, suppose that after rotating an image, we need to find the pixel value at the location (6.7, 3.2). The problem with this is that there is no such thing as fractional pixel locations.

bilinear interpolation

Digitally Reconstructed Radiograph (DRR)