What is JPEG?

JPEG (Joint Photographic Experts Group) is one of the most popular file formats for storing 8 and 24 bit continuos tone grayscale and color images.

The high efficient JPEG compression modes are:

Non-differential Huffman coding

Differential Huffman coding

Non-differential arithmetic coding.

Differential arithmetic coding.

Every class of compression method is based on the DCT (Discrete Cosine Transform) with un-correlated co-efficients. Each co-efficient can be treated independently without loss of compression efficiency. JPEG transforms the RGB data of the image into a YUV coordinate system, where the new coordinates can be evaluated in the following way:

Y = 0.3R + 0.6G + 0.1 B

V = R-Y

U = B-Y

A JPEG file can be navigated with the following markers.

Marker	Code	Length	Category
SOF0	0xFFC0	Variable	Baseline DCT
SOF1	0xFFC1	Variable	Extended sequential DCT
SOF2	0xFFC2	Variable	Progressive DCT
SOF3	0xFFC3	Variable	Sequential losses
SOF5	0xFFC5	Variable	Differential sequential DCT
SOF6	0xFFC6	Variable	Differential progressive DCT
SOF7	0xFFC7	Variable	Differential losses
SOF9	0xFFC9	Variable	Extended sequential DCT
SOF10	0xFFCA	Variable	Progressive DCT
SOF11	0xFFCB	Variable	Sequential losses
SOF13	0xFFCD	Variable	Differential sequential DCT
SOF14	0xFFCE	Variable	Differential progressive DCT
SOF15	0xFFCF	Variable	Differential losses
APPn	0xFFE0-0xFFEF	Variable	Application markers
COM	0xFFFE	Variable	Comment
DAC	0xFFCC	Variable	Arithmetic conditioning table(s)
DHP	0xFFDE	Variable	Define hierarchical progression
DHT	0xFFC4	Variable	Huffman table(s)
DNL	0xFFDC	4	Number of lines.
DQT	0xFFDB	Variable	Quantization tables(s)
DRI	0xFFDD	4	Restart interval
EOI	0xFFD9	0	End of image
EXP	0xFFDF	3	Expand reference image(s)
JPG	0xFFC8		Reserved
JPGn	0xFFF0-0xFFFD		Reserved
RES	0xFF02-0xFFBF		Reserved
RSTn	0xFFD0-0xFFD7		Restart with module 8 counter
SOI	0xFFD8	0	Start of image
SOS	0xFFDA	0	Start of scan
TEM	0xFF01	0	For temporary use

JPEG divides the compression sequence into frames and scans. Frames define the basic attributes of the image (size, number of components and coding technique).

The frame header (SOFn) structure is as follows:

Tag		Symbol	Size in bits
Frame header length		Lf	16
Sample precision		P	8
Number of lines		Y	16
Number of samples per line		X	16
Number of components in frame		Nf	8
Frame components (1 ... Nf)	Component id.	Ci	8
	Horizontal sampling factor	Hi	4
	Vertical sampling factor	Vi	4
	Quantization table destination selector	Tqi	8

Scans: The scan header(SOS) structure:

Tag		Symbol	Size in bits
Scan header length		Ls	16
Number of components in scan		Ns	8
Scancomponents (1 ... Ns)	Scan component selector	Csk	8
	DC entropy coding table selector	Tdk	4
	AC entropy coding table selector	Tak	4
Start of spectral selection of predictor selection		Ss	8
End of spectral selection		Se	8
Successive approximation bit position high		Ah	4
Successive approximation bit position low or point transform	Quantization table destination selector	Al	4

Using custom Huffman code table significantly improves the efficiency of coding. Once a Huffman table is defined, it may be used for subsequent images. The DHT marker segment structure:

Tag		Symbol	Size in bits
Huffman table definition length		Lh	16
For each Huffman table	Table class	Tc	4
	Huffman table identifier	Th	4
	Number of Huffman codes of length (1..16)	Li	8
	Value associated with each Huffman codes (1..16) (1..Li)	Lij	8

The DAC marker segment structure:

Tag		Symbol	Size in bits
Arithmetic coding conditioning table length		La	16
For each table	Table class	Tc	4
	Arithmetic coding condition identifier	Ta	4
	Conditioning table value	Cs	8

The DQT marker segment structure

Tag		Symbol	Size in bits
Quantization table definition length		Lq	16
For each table	Quantimization table element precision	Pq	4
	Quantimization table identifier	Tq	4
	Quantimization table element (0..63)	Qk	8 or 16

JPEG has a mechanism for dividing the compressed data into independently decodable segments called restart intervals. If restart is enabled in an RSTm marker, an RST marker must be placed between the entropy coded segments for each restart interval. The DRI marker segment structure:

Tag	Symbol	Size in bits
Define restart interval segment length	Lr	16
Restart interval	Ri	16