Jpeg file structure layed out simple and easy
by bkenwright@xbdev.net
Complete listing of all the possible markers with a jpeg
file.
UNI-MARKERS
SOI (0xffd8) Start Of Image
EOI (0xffd9) End Of Image
RST0 to RST7 (0xffd0
– 0xffd7) Restart Markers (placed
within the SOS block).
TEM (0xff01) Temporary for arithmetic coding.
DATA-MARKERS
SOF0 (0xffc0) Start Of Frame, baseline.
SOF1 (0xffc1) Start Of Frame, extended sequential
SOF2 (0xffc2) Start Of Frame, progressive
SOF3 (0xffc3) Start Of Frame, lossless
DHT (0xffc4) Define Huffman Table
SOF5 (0xffc5) Start of frame, differential
sequential
SOF6 (0xffc6) Start of frame, differential
progressive.
SOF7 (0xffc7) Start of frame, differential
lossless.
JPG (0xffc8) reserved.
SOF9 (0xffc9) Start of frame, extended sequential,
arithmetic coding.
SOF10 (0xffca) Start of frame, progressive,
arithmetic coding.
SOF11 (0xffcb) Start of frame, lossless, arithmetic
coding.
DAC (0xffcc) Define Arithmetic Coding Conditions.
SOF13 (0xffcd) Start of frame, differential
sequential, arithmetic coding.
SOF14, (0xffce) Start of frame, differential
progressive, arithmetic coding.
SOF15, (0xffcf) Start of frame, differential
lossless, arithmetic coding.
SOS (0xffda) Start Of Scan
DQT (0xffdb) Define Quantization Table
DNL (0xffdc) Define number of lines
DRI (0xffdd) Define restart inverval.
DHP (0xffde) Define hierarchial progression.
EXP (0xffdf) Expand reference components
APP0 to APP15 (0xffe0–0xffef)
Application-specific data
COM (0xfffe) Comment
JPG0 to JPG13 (0xfff0-0xfffd) Reserved
RES (0xff02-0xffbf) Reserved.
Sequential Files
(90 percent of jpeg files use this format, other 9 percent
progressive, and 1 percent other).
0xffd8 (SOI)
-------------------------------------------------------
0xffe0-0xffef (APP) – Application that saves the file may
embed information…e.g. adobe photoshop etc, it can be ignored.
(Length
– variable).
--------------------------------------------------------
0xffdb (DQT) – Define
Quantization Table.
-------------DQT Format Details----------------------
Parameter Size
(bits) Values Description
DQT 16 0xffdb Define quantisation table marker.
Lq 16 65n + 2 Quantization Table Length
Pq 4 0 Quantization Table Precision
Tq 4 0-3 Quantization Table Identifier
Qk 8 1-255 Quantization Elements
e.g. Q0, Q1, Q2… Q62, Q63.
(each quantisation table has its own marker, so you could
have 2 quantization markers which one is for the Y component, and the other for
the Cb and Cr components).
------------------------------------------------------
0xffc0 (SOF) – Start Of Frame
------------SOF Format Details----------------------
Parameter Size
(bits) Values Description
SOF 16 0xffc0 Start Of Frame
Lf 16 3Nf+8 Frame header length
P 8 8 Sample precision
Y 16 0-65535 Number of lines
X 16 1-65535 Samples per line
Nf 8 1-255 Number of image components (e.g.
3 for Y, U and V).
--------------Repeats for the number of components (e.g. Nf)-----------------
Ci 8 0-255 Component identifier
Hi 4 1-4 Horizontal Sampling Factor
Vi 4 1-4 Vertical Sampling Factor
Tqi 8 0-3 Quantization Table Selector.
-------------------------------------------------------
0xffc4 (DHT) – Define Huffman Table
-------------DHT Format Details---------------------
Parameter Size
(bits) Values Description
DHT 16 0xffc4 Define Huffman Table Marker
Lh 16 * Huffman Table Definition Length
Tc 4 0,1 Table Class; 0=DC 1=AC
Th 4 0,1 Huffman Table Identifier
Li 8 0-255 Huffman Code Length
(repeats 16 times, for each of how many 1bit codes, how many
2 bit cods… how many 16 bit codes).
Vij 8 0-255 Huffman Code Value
(repeats x times… where x is how many codes was determined
for Li, e.g. if there was only 4, two-bit codes and 7 three-bit codes, we would
have a sequence of 11 bytes here, each byte representing the equivalent
bit-code.
------------------------------------------------------
0xffd8 (SOS) – Start Of Scan
------------------------------------------------------
Parameter Size
(bits) Values Description
SOS 16 0xffd8 Start Of Scan
Ls 16 2Ns + 6 Scan header length
Ns 8 1-4 Number of image components
Csj 8 0-255 Scan Component Selector
Tdj 4 0-1 DC Coding Table Selector
Taj 4 0-1 AC Coding Table Selector
Ss 8 0 Start of spectral selection
Se 8 63 End of spectral selection
Ah 4 0 Successive Approximation Bit High
Ai 4 0 Successive Approximation Bit Low
-------------------------------------------------------
~~~~~ After the SOS Header Information follows the encoded
data which is in the following format.~~~~~~~
[dc huf][value] [ac
huf][value] [ac huf] [value] [ac huf] [value]…..
repeats for each component.
-------------------------------------------------------
Decoding is as follows.
Get DC Huf Value from the bits… read each one, one at a time
until we have a match. This decodes to
how many bits we read next to get our actual value that we want.
After we have obtained this value we start again.
Get AC Huf value this time, so we read in the bits one at a
time until we get a match in our huf codes.
This decodes to “two” parts, e.g. 00101 decodes to 0xAE, the A stands
for how many zeros in our 64 element array, and the E stands for the number of
bytes we should read in next to get our wanted value which will follow these
zeros. (its based on run-length encoding).
We keep repeating the AC huf coding reading, until we reach
a ac-huf value that decodes to 0x00 which mean the rest are zero, or we reach
64, which indicated that we have reached the end of this block.
Data is stored in blocks of 8x8 e.g. a sequence of 64
elements. We must extract the elements
from the coded data an put them into the 64 element array, and we’ll de-zigzag
it and inverse cosine transform it and shift is by 128, then convert the YUV to
RGB….and wow we will have an image!
I didn’t’ say it was going to be easy to decode a jpeg file.
---------------------------------------------------------
0xffd9 (EOI)
~Note to the wise~
The Jpeg file could contain APP markers or COM markers which
could effectively be ignored, just read in the header size that follows the
marker id, e.g. 0xfffe, and read to its end.
