The DNG-specification says: "If BitsPerSample is not equal to 8 or 16 or 32, then the bits must be packed into bytes using the TIFF default FillOrder of 1 (big-endian), even if the TIFF file itself uses little-endian byte order."
Assume I want to encode two 12-bits values r and g (where r11 is the MSB of r and r00 is the LSB). How would the bytes look like? (in my example the bits of each byte are listed as usual from MSB (#7) to LSB (#0))
Big endian file ('MM') and FillOrder = 1
0x0000: r11 r10 r09 r08 r07 r06 r05 r04
0x0001: r03 r02 r01 r00 g11 g10 g09 g08
0x0002: g07 g06 g05 g04 g03 g02 g01 g00
Little endian file ('II') and FillOrder = 1
0x0000: r07 r06 r05 r04 r03 r02 r01 r00
0x0001: r11 r10 r09 r08 g03 g02 g01 g00
0x0002: g11 g10 g09 g08 g07 g06 g05 g04
In previous specifications (TIFF 4.0 - 6.0) FillOrder itself has never been described very well for images with more than 1 bits per sample. So just for my understanding: If FillOrder = 2 would have been allowed in DNG, I'd assume the following layouts:
Big endian file ('MM') and FillOrder = 2
0x0000: r11 r10 r09 r08 r07 r06 r05 r04
0x0001: g11 g10 g09 g08 r03 r02 r01 r00
0x0002: g07 g06 g05 g04 g03 g02 g01 g00
Little endian file ('II') and FillOrder = 2
0x0000: r07 r06 r05 r04 r03 r02 r01 r00
0x0001: g03 g02 g01 g00 r11 r10 r09 r08
0x0002: g11 g10 g09 g08 g07 g06 g05 g04
To put it short (writer's perspective):
- The file endianess describes how to cut values into octets (bytes) and how to order them.
- The FillOrder describes the order in which fragments of consecutive values are to be joined into a single octet.
Is my interpretation correct?
Bonus question: Is the FillOrder-Field forbidden in DNG? What is a compatibility-oriented reader assumed to do, when it encounters a FillOrder=2 - use it as given or ignore it and enforce FillOrder=1?
AdChoices