When barrel distortion / radial distortion occurs, pixels at the edge map to larger real world distance than pixels at the center. In order to undistort this, pixels near the center must be "compressed".
The non-uniform distribution of pixels mean that the final undistorted image must have lower resolution than the original image.
However, when using getOptimalNewCameraMatrix with alpha=0, the final image ROI is the same as the original image.
How is that possible?