1. A 5x5 pixel (or larger) window W within the active sensor array is chosen. The term 'pixel' here is taken to mean a group of 4 Bayer mosaic channels (R, Gr, Gb, B). The location and size of the window chosen is implementation defined, and should be chosen to provide a green split estimate that is both representative of the entire image for this camera sensor, and can be calculated quickly.
2. The arithmetic mean of the green channels from the red rows (mean_Gr) within W is computed.
3. The arithmetic mean of the green channels from the blue rows (mean_Gb) within W is computed.
4. The maximum ratio R of the two means is computed as follows: R = max((mean_Gr + 1)/(mean_Gb + 1), (mean_Gb + 1)/(mean_Gr + 1))

• R < 1.03 is a negligible split (<3% divergence).
• 1.20 <= R >= 1.03 will require some software correction to avoid demosaic errors (3-20% divergence).
• R > 1.20 will require strong software correction to produce a usuable image (>20% divergence).

The worst-case divergence between Bayer green channels.

This value is an estimate of the worst case split between the Bayer green channels in the red and blue rows in the sensor color filter array.

The green split is calculated as follows:

The ratio R is the green split divergence reported for this property, which represents how much the green channels differ in the mosaic pattern. This value is typically used to determine the treatment of the green mosaic channels when demosaicing.

The green split value can be roughly interpreted as follows:

Range of valid values:

>= 0

Optional - This value may be null on some devices.

[Android Documentation]