Remap on NVIDIA GPU
Updated:
Remap on CUDA moves each pixel to a new position according to a pair of 2D offset maps, so a single fast GPU pass can fix geometric distortions, rotate an image, or correct colour differences between channels — per channel or all together.
Applications
- Radial and tangential undistortion; chromatic aberration removal
- Rotation by arbitrary angles; dewarping for cylindrical or spherical projections
- Affine and perspective transforms; arbitrary projections via 2D maps
How it works
- Capture calibration images with a chessboard pattern
- Compute the distortion coefficients and the image-centre position
- Compute pixel offsets from the expected positions (with predefined X/Y steps)
- Build two maps — X offsets and Y offsets
- Remap reads the two offset tables and repositions pixels; new pixels are interpolated (linear, cubic or Lanczos)
Remap is a module of our CUDA Image & Video Processing SDK and can run as one stage of a real-time GPU pipeline.