http://gvsoft.no-ip.org/exif/

rev. 1.1 2005-06-03
rev. 1.0 2005-04-01

• 1. What is Exif?
• 2. JPEG format and Marker
• 3. EXIF marker
• 4. Exif data structure
  • 4.1 TIFF Header
  • 4.2 IFD : Image File Directory
  • 4.3 IFD data structure example
  • 4.4 Thumbnail image
    • 4.4.1 JPEG format thumbnail
    • 4.4.2 TIFF format thumbnail
• Tag number used by Exif / TIFF
  • IFD0 (IFD of main image) section
  • Exif SubIFD section
  • IFD1 (IFD of thumbnail image) section
  • Misc Tags

Basic format of Marker is below.
0xFF+Marker Number (1 byte) + Data [ Size (2 bytes) + Info (n bytes) ]

The 'Data' contains Data Size descriptor and Information, Size (2 Bytes) has 'Motorola' byte align. The data size is the size of the total Data: Size + Info ( value = 2 + n).

Example: FF xx - 00 0C - 01 02 03 04 05 06 07 08 09 0A

It means this Marker (0xFFxx) has 0x000C (12) bytes of data. The data '12 bytes' starts with 'Data size' 2 bytes, followed by 10 bytes: '01 02 03 04 05 06 07 08 09 0A'.

In JPEG format, some of Markers describe data, then SOS (Start Of Stream) marker placed. After the SOS marker, JPEG image stream starts and terminated by EOI  (End Of Image) marker. 

For inserting data Exif uses the Application Marker APP1 (0xFFE1) marker to avoid a conflict with the JFIF format. Every Exif file format starts from this marker. 

After the 'SSSS', APP1 data starts. This part is explained in the next section.

After the APP1 marker area other JPEG markers follow. 

First 2 bytes defines byte align of TIFF data. 0x4949='I I' is 'Intel' type byte align, 0x4d4d='MM' is 'Motorola' type byte align. For example, decimal value '305.419.896' is noted as hexadecimal 0x12345678. If Motorola aligned, it is noted as 0x12345678, if it's Intel aligned it is noted as 0x78563412. A lot of digital camera uses Intel align, but not all. Therefore when we need the value of Exif data, we MUST check byte align field every time. Though JPEG data uses Motorola align only, Exif allows both alignments. I can't understand why Exif didn't fix a byte align to Motorola.

Next 2 bytes always have the value of 0x002A.

The last 4 bytes of TIFF header is the offset to the first IFD (Image File Directory, described in next chapter). Includes this offset, all the offset value used in TIFF format counts offset bytes from the first byte of the TIFF header ('I I' or 'MM'). Usually the first IFD starts immediately next to TIFF header, so this offset has often the value '0x00000008'. 

Next to TIFF header, there is the first IFD : Image File Directory - IFD-0. It contains image information data.

The first 2 bytes ('eeee') contains the number of directory entries in this IFD.

Then directory entries (12 bytes per entry) follows.

'tttt'	2 bytes	is the Tag number, this shows a kind of data.
'ffff'	2 bytes	is data format.
'nnnnnnnn'	4 bytes	is number of components.
'dddddddd'	4 bytes	contains a data value or offset to data value.

After last directory entry, there is a 4bytes of data ('oooooooo'), it has the offset to next IFD. If its value is '0x00000000', it means this is the last IFD and there is no linked IFD.

Data format ('ffff') is defined as see table.

Value	Format	Bytes/component
1	unsigned byte	1
2	ascii strings	1
3	unsigned short	2
4	unsigned long	4
5	unsigned rational	8
6	signed byte	1
7	undefined	1
8	signed short	2
9	signed long	4
10	signed rational	8
11	single float	4
12	double float	8

'rational' means a fractional value, it contains 2-signed/unsigned long integer values, and the first represents the numerator, and the second, the denominator.

In case of Extended Exif format (Exif2.1/DCF), Exif SubIFD contains special Tag Exif Interoperability Offset (0xa005). It also points to the Interoperability IFD. By the DCF specification, this tag is mandatory and both SubIFD (main image IFD) and IFD1 (thumbnail image IFD) may have Interoperability IFD. In usual, only main image have this tag.

And some of digital camera uses IFD data format for Makernote; Maker-specific magic data area. It's very hard to decide whether makernote is IFD format or not, be careful to coding program. See Appendix for information of Makernote.

0000: 49 49 2A 00 08 00 00 00-02 00 1A 01 05 00 01 00
0010: 00 00 26 00 00 00 69 87-04 00 01 00 00 00 11 02
0020: 00 00 40 00 00 00 48 00-00 00 01 00 00 00

If thumbnail uses no compression and PhotometricInterpretation (0x0106) Tag in IFD1 has a value '2', thumbnail uses RGB format. In that case, you can see thumbnail image by simply copy data to computer's RGB format (such as BMP format, or copy to VRAM directory). Kodak DC-210/220/260 uses this format. Be mention that at TIFF stores pixel data as 'RGB' order, but BMP stores 'BGR' order.
If that tag has a value '6', thumbnail uses YCbCr format. If you want to see thumbnail, you must convert it to RGB. Ricoh RDC4200/4300, Fuji DS-7/300 and DX-5/7/9 use this format (newer RDC5000/MX-X00 series use JPEG). Next section is brief description to conversion of Fuji DS's thumbnail. For more details, refer to TIFF6.0 specification.

At DX-5/7/9, YCbCrSubsampling(0x0212) has values of '2,1', PlanarConfiguration(0x011c) has a value '1'. So the data align of this image is below.

Y(0,0),Y(1,0),Cb(0,0),Cr(0,0), Y(2,0),Y(3,0),Cb(2,0),Cr(3.0), Y(4,0),Y(5,0),Cb(4,0),Cr(4,0). . . .

The numeric in parenthesis is pixel coordinates. DX series' YCbCrCoefficients(0x0211) has values '0.299/0.587/0.114', ReferenceBlackWhite(0x0214) has values '0,255,128,255,128,255'. Therefore to convert from Y/Cb/Cr to RGB is;

B(0,0)=(Cb-128)*(2-0.114*2)+Y(0,0)
R(0,0)=(Cr-128)*(2-0.299*2)+Y(0,0)
G(0,0)=(Y(0,0)-0.114*B(0,0)-0.299*R(0,0))/0.587

Horizontal subsampling is a value '2', so you can calculate B(1,0)/R(1,0)/G(1,0) by using the Y(1,0) and Cr(0,0)/Cb(0,0). Repeat this conversion by value of ImageWidth(0x0100) and ImageLength(0x0101).