The current algorithm uses numerical pixel-splitting, which is accurate but slow. (8 pixel corner h k l evaluations per pixel at minimum)
When the RS is oversampled anyway, we could use a simple one h k l per pixel approach, which should improve performance substantially.
The current algorithm uses numerical pixel-splitting, which is accurate but slow. (8 pixel corner h k l evaluations per pixel at minimum)
When the RS is oversampled anyway, we could use a simple one h k l per pixel approach, which should improve performance substantially.