CODEWITHSUNDEEP
parsingvideoh.264
Pobbel
Pobbel

Reputation: 179

H.264 NALU Byte Alignment

I am trying to get my head around the H.264 NALU headers in the following data stored in a mov container.

Example from file:

00 00 00 02 09 30 00 00 00 0E 06 01 09 00 02 08
24 68 00 00 03 00 01 80 00 00 2B 08 21 9A 01 01
64 47 D4 B2 5C 45 76 DA 72 E4 3B F3 AE A9 56 91
B2 3F FE CE 87 1A 48 13 14 A9 E0 12 C8 AD E9 22
...

So far I have assumed that the bit-stream is not byte aligned due to the start code sequence offset to the left by one bit:

0x00 0x00 0x00 0x02 -> 00000000 00000000 00000000 00000010

So I have shifted the these and subsequent bytes to the right one bit which results in the following start sequence code and header bits for the first header:

0000000 00000000 00000000 00000001 [0 00 00100]

However I am coming unstuck when I reach following byte sequence in the example:

0x00 0x00 0x00 0x0E

I am assuming it is another start sequence code but with a different byte alignment.

00000000 00000000 00000000 00001110 00000110 00000001 00001001 00000000

After byte alignment I am getting the following header byte:

00000 00000000 00000000 00000001 [1 10 00000]

The first bit in the header (the forbidden_zero_bit) is non-zero which violates the rule that it must be zero

Where am I tripping up?

Am I making the wrong assumptions here?

Upvotes: 0

Views: 1935

Answers (2)

nobody555
nobody555

Reputation: 2374

As was already answered MOV-container (or MP4) doesn't use Annex B encoding with start codes. It use MP4-style encoding where NALs are prefixed with NALUnitLength field. This field can be of different size (and that size signaled somewhere else in container) but usually it is 4 bytes. In your case NALUnitLength is probably 4 bytes and 3 NALs from you dump have sizes of: 2-bytes (00 00 00 02), 14-bytes (00 00 00 0E) and 11016-bytes (00 00 2B 08).

Upvotes: 1

Roman Ryltsov
Roman Ryltsov

Reputation: 69652

Start codes are used in "Byte stream format" (H.264 Annex B) and are byte aligned themselves. Decoder is supposed to identify start code by checking byte sequences, without bit shifts.

MOV, MP4 containers don't use start codes, however they have their own structure (atoms, boxes) with parameter set NAL units, without prefixes, in sample description atoms and then data itself separately again as original NAL units.

What you quoted is presumably a fragment of MOV atoms which correspond to file structure bytes and not NAL units.

Upvotes: 1

Related Questions