Bit masks

When masks are used

Masks apply when the compression is BI_BITFIELDS (3), which gives red, green and blue masks, or BI_ALPHABITFIELDS (6), which adds an alpha mask. Both are only valid at 16 and 32 bits per pixel. BI_ALPHABITFIELDS came from Windows CE and is rarely seen elsewhere.

With BI_RGB, no masks are stored and the fixed defaults below apply.

Where they are stored

Where the masks sit depends on the version of the DIB header:

Each mask is a little-endian 32-bit number. In the example above, the red mask 00 F8 00 00 is 0x0000F800.

Reading a channel

To read a channel, AND the pixel with its mask and shift the result right by the number of zero bits below the mask. The result runs from 0 to the mask’s maximum, such as 31 for a 5-bit mask, so scale it to 0 to 255 for display:

const shift = trailingZeros(mask);
const max = mask >>> shift;
const value = (pixel & mask) >>> shift;
const channel = Math.round(value * 255 / max);

The set bits of each mask must be one unbroken run, and masks must not overlap. A mask of zero means the channel is absent.

Default and common masks

The defaults are used with BI_RGB. For a 32-bit BI_RGB image the top byte is officially unused, though some programs store alpha there; a reader cannot tell for certain, so writers that want alpha should use BI_BITFIELDS with a V4 or V5 header and an alpha mask.