A distortion model defines a coordinate mapping between an ideal rectilinear image and the distorted image captured by the lens.

In practice, undistortion usually works like this:

1. For each pixel location in the corrected output image, the algorithm computes where that point would fall in the original distorted image using the calibrated parameters (focal length, principal point, radial and tangential distortion terms).
2. That mapped location is often not exactly on a source pixel center, so the pixel value must be estimated by interpolation from nearby pixels.
3. Because distortion changes local magnification, one output pixel may correspond to a differently sized area in the input image. A good resampling method should account for this so brightness/energy is preserved as well as possible.

So the model itself is not “removing distortion” directly; it supplies the geometric transformation. The actual correction comes from resampling the recorded image onto a new rectangular grid using that mapping.

This is why calibration parameters matter: they determine the exact warp from ideal image coordinates to distorted sensor coordinates.