White balance correction is usually done by scaling color channels so a known neutral area becomes neutral. In RGB, if you can identify a patch that should be white or gray, measure its R, G, and B values and multiply each channel by a factor that makes those values equal. Example: if a neutral sample reads (250, 245, 255), the gains would be about 255/250, 255/245, and 255/255, so after scaling it becomes neutral.

This works best when the image data is linear, because the same channel ratios then apply across the tonal range. In practice, camera processing and color spaces can complicate the exact math, but the principle is the same: apply per-channel gains.

Without a known neutral reference—or some human judgment about what in the scene should be neutral—you generally cannot determine the “correct” white balance from RGB values alone. The image data by itself is ambiguous.

For YUV, white balance is still fundamentally a color-channel correction problem, so it is typically easier to think of it in RGB (or the camera’s native sensor data), apply the channel gains there, then convert back if needed.