.JPG File Format – Markers and segments

JPEG image (JFIF and Exif)

A JPEG file is a run of segments, and each segment begins with a marker. The markers and what each segment holds are defined in Annex B of ITU-T T.81 (ISO/IEC 10918-1). Every number in a JPEG file is big-endian.

Markers

A marker is two bytes: FF, then a code that is neither 00 nor FF. Any marker may be preceded by any number of extra FF bytes, called fill bytes.

CodeNameWhat it isSegment body
C0SOF0Frame header: baseline DCTLength and frame header
C1SOF1Frame header: extended sequential DCTAs SOF0
C2SOF2Frame header: progressive DCTAs SOF0
C3SOF3Frame header: lossless (sequential)As SOF0
C4DHTDefine Huffman tablesLength and Huffman tables
C5–C7SOF5–SOF7Frame headers: differential (hierarchical), Huffman codingAs SOF0
C8JPGReserved for JPEG extensionsNot defined
C9–CBSOF9–SOF11Frame headers: extended, progressive and lossless, arithmetic codingAs SOF0
CCDACDefine arithmetic coding conditioningLength, then pairs of bytes
CD–CFSOF13–SOF15Frame headers: differential (hierarchical), arithmetic codingAs SOF0
D0–D7RST0–RST7Restart markers, inside entropy-coded dataNone: the marker stands alone
D8SOIStart of imageNone
D9EOIEnd of imageNone
DASOSStart of scanLength, scan header, then entropy-coded data
DBDQTDefine quantisation tablesLength and tables
DCDNLDefine number of linesLength and 2-byte line count
DDDRIDefine restart intervalLength and 2-byte count of MCUs
DEDHPDefine hierarchical progressionAs a frame header
DFEXPExpand reference componentsLength and 1 byte
E0–EFAPP0–APP15Application dataLength and data: see application segments
F0–FDJPG0–JPG13Reserved for JPEG extensionsNot defined
FECOMCommentLength and text
01TEMTemporary private use in arithmetic codingNone
02–BFRESReservedNot defined

Segments

All markers except SOI, EOI, RSTn and TEM are followed by a 2-byte length. It counts itself and the rest of the segment, but not the marker, so its smallest value is 2.

  1. 0Marker FF and a code2 bytes
  2. 2Length big-endian, including these 2 bytes2 bytes
  3. 4Body as the code sayslength − 2 bytes
A segment with a length. SOI, EOI, RSTn and TEM have only the marker.

Order in a file

The file starts with SOI and ends with EOI. Frame headers, tables and scans come between them in this order; tables can appear anywhere before the scan that first uses them, and they stay in force until replaced.

  1. …SOI2 bytes
  2. …Application data, comments, tables APPn, COM, DQT, DHT, DRIvaries
  3. …Frame header SOF0, SOF1, SOF2 or another SOFnvaries
  4. …Tables DQT, DHT, DRI, as needed before each scanvaries
  5. …Scans each one SOS and its entropy-coded data; DNL may follow the firstvaries
  6. …EOI2 bytes
The layout of a file with one frame, as T.81 gives it

A file made only of tables, with no frame, is called an abbreviated format; it is how some programs store shared tables for a set of images.

The bytes after a scan

Unlike other segments, a scan has no length that covers its data. To find its end, read the bytes after the scan header until a marker that is not a restart marker. A FF byte that belongs to the data is written FF 00; see Scan.