The big difference is that the DHT marker that contains the
Huffman table for decoding is missing. This table is fixed
for MJPEG and defined in mmreg.h, according to the OpenDML
specs.

Some reading on the subject:

- JPEG standard (markers are listed in section B.1.1.3):

http://www.w3.org/Graphics/JPEG/itu-t81.pdf

- JFIF standard (JFIF marker at page 5):

http://www.w3.org/Graphics/JPEG/jfif3.pdf

- JFIF, old AVI1 and new AVI1 markers:

http://sourceforge.net/mailarchive/message.php?msg_id=7062286

- ODML standard (MJPEG DIB described at page 26):

http://www.nomadelectronics.com/VidCap/capture using
vfw/ODMLFF2.DOC

So, in the end, you need to find the 0xFF,0xDA SOS marker in
your sample (that's where the image data starts) then write
averything up to that point to the JPEG output stream, then
write the DHT, then write the rest of the sample. The DHT is
exactly the content of the MJPGDHTSeg byte array defined
near the end of mmreg.h.

Let's assume pb points to the sample data and cb is the
sample length in bytes:

 int k = 0; while(pb[k] != (char)0xFF || pb[k+1] !=
(char)0xDA) k++;
 output(&pb[0],k);
 output(MJPGDHTSeg,sizeof(MJPGDHTSeg));
 output(&pb[k],cb-k);

If you want, you can also replace the 'AVI1' APP0 with a
'JFIF' APP0, but it is not required (APP0 = 0xFF,0xE0).

--