Significant Bits (sBIT)

• Status:   PNG Specification
• Location:   before PLTE and first IDAT
• Multiple:   no

The significant-bits chunk is used to indicate the nature of the source data in cases in which storing it in PNG form required a conversion. For example, gray pixels in medical images are often 12 bits deep, but PNG requires them to be scaled up to 16 bits for portability. Scaling the pixels does not alter the fact that they contain only 12 bits of real information, and the sBIT chunk stores this fact in a PNG file.

As with several other PNG chunks, the format of sBIT depends on the image type. Grayscale images are the simplest; sBIT then contains a single byte indicating the number of significant bits in the source data--in the preceding example, 12. For grayscale images with an alpha channel, sBIT contains 2 bytes, one for the gray channel and one for alpha; RGB images require 3 bytes, and RGBA images require 4. Palette-based images are treated like RGB except that the sBIT information refers to the palette entries, and the palette's effective sample depth is always 8, regardless of how many bits are used to index the palette. Note that the number of significant bits for any given channel must be greater than zero and less than or equal to the sample depth.

Ordinary PNG decoders need not worry about sBIT, but those that wish to recover the original image data can do so by right-shifting each image sample to leave only the number of bits indicated by sBIT. This implies that the scaling procedure used by the PNG encoder must not change the original bits; it can only append low-order bits to each sample.