jpeg processing modules for real-time decoding and …field programmable port extender (fpx) 6...

Field Programmable Port Extender (FPX) 1

JPEG Processing Modules forReal-Time Decoding and Recoding

Chris Neely

Washington University, Applied Research Lab

June 19, 2002

Supported by: NSF

http://www.arl.wustl.edu/arl/projects/fpx


Outline

• Introduction• Motivation of JPEG

Recoder Module• JPEG Encoding Concepts• FPX JPEG Recoder

• Up to N coefficients• Block Diagram• Image Results• Thumbnail size Images• Network Configuration

Hardware Synthesis


Hardware JPEG Application

• FPX Video Recoder ModuleThis module was implemented to adjust the bandwidth of MotionJPEG video midway through the network. Control cells are sent tospecify the number of coefficients to drop on eachluminance/chrominance channel (YCbCr).


Motivation for JPEG Recoding

MMXMMX

NTSC

NTSC

JPEG Compressed VideoVideo

Monitor

Network

FPX



MMXMMX

NTSC

NTSC

Video

Monitor

Network

FPX

Congestion Flashes or No Video



MMXMMX

NTSC

NTSC

Video

Monitor

Network

FPX

Recoded Video ReducesDownstream Bandwidth

Video continues to playbut with less detail

Congestion


FDCT JPEG entropy encoder

Table Specifications

Quantizer

Table Specifications

CompressedImageData

JPEG Encoder – 4 Step Process

31 24

SourceImageData

JPEG Encoder – 4 Step Process


JPEG Encoding Concepts

• First, encode the information that is the most valuableto human perception.

• Convert from RGB to a luminance chrominancerepresentation to preserve the luminance becausethat is considered valuable.

• Divide into 8x8 block computational units becausethey are small and efficient for processing.

• The human eye functions like a low pass filter, soencode low spatial frequencies with more accuracythan high frequencies.


JPEG Encoding Concepts

• The compression ratio is controlled by varying thequantization tables. Fewer/smaller coefficients afterquantization means greater compression.

• Use variable length representations for data thatassign as few bits as possible to frequently occurringdata patterns (entropy encoding).

• 0xFF Marker codes indicate the start and end ofimages.


Raw ATM cells containing RAW JPEG

Motion JPEG Input Cell Format:

0xFFE0 = Start of ImageAAL0 Cell Header

Network Data format for Motion-JPEG Video


JPEG Video Recoder Design

ByteStuffer

EntropyDecoder

Convert toThumbnail

Control CellProcessor

Bit SerialEgress

EntropyRecoder

JPEG_MODULE

OutputController


This pattern begins by mapping thethe value at the lowest spatialfrequency, the DC coefficient.

The first AC coefficient is next.

Now the Zig-Zag pattern

The last values to be coded are thosethat relate to the highest spatialfrequencies.

4.1. Follow Zig-Zag Coding SequenceData Order is a Zig-Zag Sequence

0 1 5 6 14 15 27 28

2 4 7 13 16 26 29 42

3 8 12 17 25 30 41 43

9 11 18 24 31 40 44 53

10 19 23 32 39 45 52 54

20 22 33 38 46 51 55 60

21 34 37 47 50 56 59 61

35 36 48 49 57 58 62 63

Increasing Horizontal Frequency

Incr

ea

sing

Ve

rtic

alF

req

uenc

y


The Huffman table for theVariable Length Codedescribes therun-length pairs: the zero-run and thenumber of zeros in the integer.

4.1. Follow Zig-Zag Coding SequenceHuffman Variable Length Code Tables

Run/Sz Co_len Co_wrd0/0 2 00/1 3 20/2 3 30/3 3 40/4 3 50/5 3 60/6 4 e0/7 5 1e0/8 6 3e0/9 7 7e0/A 8 fe0/B 9 1fe

Run/Sz Co_len Co_wrd0/0 4 a0/1 2 00/2 2 10/3 3 4

…F/6 16 fffaF/7 16 fffbF/8 16 fffcF/9 16 fffdF/A 16 fffe


TheVariable Length Integers can be up to 11 bits for the DC component and 10 bitsfor the AC component.

4.1. Follow Zig-Zag Coding SequenceVariable Length Integers

Integer Value-1,1

-3,-2, 2, 3-7..-4, 4..7

-15..-8, 8..15-31..-16, 16..31-63..-32, 32..63

-127..-64, 64..127-255..-128, 128..255-511..-256, 256..511

-1023..-512, 512..1023

Number of bits123456789

10

VLI Representation0, 1

00, 01, 10, 11000..011, 100..111

0000..0111, 1000..111100000..01111, 10000..11111

000000..011111, 100000..1111110000000..0111111, 1000000..1111111

00000000..01111111, 10000000..11111111000000000..011111111, 100000000..111111111

0000000000..0111111111, 1000000000..1111111111


Entropy Decoder Module

• Bit-serial shift register– Bit-serial data-in– Identifies 0xFFEx marker control codes– Removes byte stuffing (0xFF00)– If (vlc_state = ‘1’), then shifts overflow to VLC register

for lookup operation– If (vli_state = ‘1’), then shifts overflow to store VLI

based on lookup result• Variable length data output

– VLI (up to 11 bits) is reverse-sign extended to a fixed12 bit output

– VLC (up to 16 bits) is padded with zeros to a fixed 16bit output

• 4 Huffman Code Tables implemented asCombinational Logic



Huffman TableSpecifications

Combinational Logic

DataENDATA

LUM_CHR_H

RESET_L, CLK

HUFF_TBL_SEL

Serial Shift Reg.

DecoderFSM

SAMP_MODE

VLC_OUTVLI_OUTPOSITION_OUTBLOCK_NUM_OUT

SOI_OUTEOI_OUTEOB_OUTMCU_OUTVLC_OUTVLI_OUTEVEN_FIELD_L

111110110 0000 1111111110100111

VLC VLI

1 First shift register is not full yet.




Combinational Logic

DataENDATA

LUM_CHR_H

RESET_L, CLK

HUFF_TBL_SEL

Serial Shift Reg.

DecoderFSM

SAMP_MODE



111110110 0000 1111111110100111

VLC VLI

1111101100000111

AC value (Zero Run = 1, Bits = 4)

Filled first shift register, soshifting overflow to VLC




Combinational Logic

DataENDATA

LUM_CHR_H

RESET_L, CLK

HUFF_TBL_SEL

Serial Shift Reg.

DecoderFSM

SAMP_MODE



VLC VLI

1111011000001111 1

111110110 0000 1111111110100111

Doing Lookup on VLCon each shift.




Combinational Logic

DataENDATA

LUM_CHR_H

RESET_L, CLK

HUFF_TBL_SEL

Serial Shift Reg.

DecoderFSM

SAMP_MODE



VLC VLI

0000111111111010 11110110

1111101100000 1111111110100111

Found VLC = “11110110”AC value (Zero Run = 1, Bits = 4)




Combinational Logic

DataENDATA

LUM_CHR_H

RESET_L, CLK

HUFF_TBL_SEL

Serial Shift Reg.

DecoderFSM

SAMP_MODE



VLC VLI

0001111111110100 0

1111101100000 1111111110100111

Find VLI of size 4 bits.




Combinational Logic

DataENDATA

LUM_CHR_H

RESET_L, CLK

HUFF_TBL_SEL

Serial Shift Reg.

DecoderFSM

SAMP_MODE



VLC VLI

1111111110100111 0000

11111011000001111111110100111

Found VLI = “0000”,Integer = -15.


Symmetric Entropy FSM

(hit = '1' & SSSS > 0)K += R

block_num++

block_num++

LUM_DC_VLC

IDLE

LUM_DC_VLI

LUM_AC_VLC LUM_AC_VLI

(hit = '1' & SSSS >0)

(vli_done ='1')(hit = '1' & SSSS = 0)

(hit = '1' & SSSS > 0)K += R

(R = 0)?

(hit = '1' & SSSS = 0)

Y

N(R = 15)?

Y

N

K += 16

(vli_done = '1')

(K = 63)?

K++

N

Y

CHR_DC_VLC CHR_DC_VLI

CHR_AC_VLC CHR_AC_VLI

(hit = '1' & SSSS >0)

(vli_done ='1')(hit = '1' & SSSS = 0)

(R = 0)?

(hit = '1' & SSSS = 0)

Y

N(R = 15)?

Y

N

K += 16

(vli_done = '1')

(K = 63)?

K++

N

Y

(SOI = '1')

End of ImageMarker

(EOI = '1')

Decoded all MCULuminance

Blocks?

Y

N

Decoded both MCUChrominance

Blocks?

Y

N

block_num++

block_num++

Start of ImageMarker

Error

(error_h = '1')

mcu_blip <= '1'

// Build VLC and Do Huffman Table lookupcodeword <= code_word(15 downto 0) & buf_val(15)code_length++

1* 2*

1*

2*

2*

1*

2*1*

2* =

1* =

// Build VLI based on the SSSS saved from Huffman lookupinteger_val <= integer_val(8 downto 0) & buf_val(15)integer_length++

LumDC_VLCLumDC_VLI

LumAC_VLC

LumAC_VLI

ChrDC_VLC

ChrDC_VLI

ChrAC_VLC

ChrAC_VLI


Entropy Recoder Module

DataENDATA

RESET_L, CLK

Shift Register (32 bits data with 32 bits overflow)

SOI_INEOI_IN

EOB_INLUM_CHR_H

Position_KN_coeff_reqVLC_ReadyVLC_valueVLI_ReadyVLI_value

EVEN_FIELD_LHUF_TBL_SEL

Recoder FSM shift_ptr

11110110

Found VLC = “11110110”AC value (Zero Run = 1, Bits = 4)

Is the coefficient position <= N?


Entropy Recoder Module

DataENDATA

RESET_L, CLK

Shift Register (32 bits data with 32 bits overflow)

SOI_INEOI_IN

EOB_INLUM_CHR_H

Position_KN_coeff_reqVLC_ReadyVLC_valueVLI_ReadyVLI_value

EVEN_FIELD_LHUF_TBL_SEL

Recoder FSM shift_ptr

111101100000

Found VLI = “0000”,Integer = -15.

Is the coefficient position <= N?


Perform Byte Stuffing

Byte Shifting Pipeline– Shift Most significant byte to

check for 0xFF’s– Stuff 0x00 if 0xFF is found that

is not a marker code

Data_out

Wr_en

Data

RecodedEntropy Data

RegA32 bits

Byte Stuffer

RegB32 bits

RegC32 bits

Shift_Cntr

Load_CntrFIFO_empty Byte Stuffer FSM






Data_out

Wr_en

Data

RecodedEntropy Data

RegA32 bits

Byte Stuffer

RegB32 bits

RegC32 bits

Shift_Cntr


Looking for 0xFF…






Data_out

Wr_en

Data

RecodedEntropy Data

RegA32 bits

Byte Stuffer

RegB32 bits

RegC32 bits

Shift_Cntr


Found 0xFF!

Do an extra shift in order to stuffwith zeros (0xFF00).


Format INPUT_CELLS.DAT

Simulation Testbench


Simulation Waveforms – Finding VLCs and VLIs

Matched VLC

Stored VLI


Reformat SW_CELLSOUT.DAT


Simulation JPEG Output – DC Values


Recoded for Greater Compression

(Normal video)

17.14 KB, 0.88 bits-per-pixelup to 8 Y, 8 Cb, 8 Cr

11.78 KB, 0.61 bits-per-pixel



(Normal video)


up to 4 Y, 4 Cb, 4 Cr




(Normal video)






(Normal video)





Convert to “Thumbnail”

• 4 Pipeline Conversion Stages– Determine VLI size and

convert to 12 bit signedinteger

– Add to 12 bit DPCM(DC running sum)

– Denormalize (convert to 8 bitunsigned integer)

– Average Blocks Horizontally

Data_out

DataEn_out

DC_VLI_ready

VLI_data

Stage1:VLI_size

Convert to Thumbnail

Stage2:Add_DPCM

Stage3:Denormalize

Stage4:Average Luma

Lum_chr_h








Data_out

DataEn_out

DC_VLI_ready

VLI_data

Stage1:VLI_size


Stage2:Add_DPCM

Stage3:Denormalize

Stage4:Average Luma

Lum_chr_h








Data_out

DataEn_out

DC_VLI_ready

VLI_data

Stage1:VLI_size


Stage2:Add_DPCM

Stage3:Denormalize

Stage4:Average Luma

Lum_chr_h


Sample Baseline Bitmap Output

Created by Image Editing Software(Corel PhotoPAINT)

JPEG Decoder Simulation Result(flipped and w/ header)


Sample MMX Bitmap Output

JPEG Decoder Simulation Result


Results

• How fast does it go?– 27.443 MHz, RAD_CLKB

• How much logic / memory does it require?– 22 logic levels– 45 % BLOCKRAMs– 12% SLICEs



MMXMMX

NTSC

NTSC

Video

Monitor

Network

FPX

NCHARGE


Video Recoder NCHARGE Module Listing

NCHARGE


NCHARGE is used from a web browser to configure the module.Module ID: 60 (0x3C)Opcode: 0x04 Manually specify the number of coefficients per color component (N1xY, N2xCr, N3xCb)Opcode: 0x10 Preset to blocky color (1xY, 1xCr, 1xCb)Opcode: 0x12 Preset to blocky grayscale (1xY, 0xCr, 0xCb)

Video Recoder NCHARGE Module Listing

NCHARGE Listing File:<module>Video Recoder</module><input_opcodes>0x04,A,3,Lum,Cb,Cr,0x10,B,0,0x12,C,0,</input_opcodes><output_opcodes>0x05, A,3,Lum,Cb,Cr,0x11,B,0,0x13,C,0,</output_opcodes><fields>Lum,x,1,22,1,16,Cb,x,1,14,1,8,Cr,x,1,6,1,0</fields><help>A Manual set coefficientsB Preset blocky colorC Preset blocky grayscale</help>


References

• Wallace, Gregory K. "The JPEG Still Picture Compression Standard"Reprinted from IEEE Transactions on Consumer Electronics.December 1991.

• Pennebaker and Mitchell. JPEG: Still Image Data CompressionStandard. Van Nostrand Reinhold. 1993.

• Neely and Bogale "Motion JPEG Decoder for MMX ImageProcessing Applications." Original project website 2001.http://userfs.cec.wustl.edu/~cen1/mmx_jpeg.htm

• Braun, Lockwood, and Waldvogel. "Layered Protocol Wrappers forInternet Packet Processing in Reconfigurable Hardware."Washington University in Saint Louis. July 2001.

• "Discrete Cosine Transform Tutorial." HTML page accessed July 10,2001.http://ce597n.www.ecn.purdue.edu/CE597N/1997F/students/aaron.r.bouillet.1/project/Application



MMX JPEG Format

• Half-size images of NTSC fields(even or odd scan lines) areencoded as individual JPEGimages (60 frames/sec)

• 0xFFxx Marker Codes identify theimages:

0xFFE0 = Start of Even Field0xFFE1 = Start of Odd Field0xFFE2 = End of Field

• Image Resolution: 640x248


Monitor

Video

Monitor

JavaApplet

CS535 Project Logical Connection Diagram

ATM

FPXMMXNTSC


Connections: MMX Video station to the FPX VR

Camera

Monitor

JPEG(0/101)

NCHARGE

MMXFPX

NTSC

NTSC

JPEG(0/101)

Control(0/35)

Contro

l(0

/142

) VideoRecoder

WUGS

Control

(0/1

52)


CS535 “Thumbnail” Image Decoder

Dequantize& Average

Entropydecoder

TableSpecifications

CompressedImage Data

640x248pixels

VCI = 0/101

UDP OutputController

1240 Byte

UDP Packet

AAL0HeaderRAWJPEG

AAL0HeaderRAWJPEG

AAL0Header

RAWJPEG

DCValues

UDP Wrappers

TableSpecifications

AAL0HeaderRAWJPEG

AAL0HeaderRAWJPEG

ATM

OTHERATM

8-bit GrayscaleImage Data

(40x31 pixels)1240 Bytes/image

Control Cell ProcUpdatePacing


The conversion from RGB to YCbCr uses the following equations:

24-bit RGB (Original)

8-bit Luminance (Y) 8-bit Chrominance Blue (Cb) 8-bit Chrominance Red (Cr)

1. Separate the Color Channels1.1. Convert the Color Channels

Y = ( LumaRed* R + LumaGreen* G + LumaBlue*B )Cb = ( B - Y ) / ( 2 - 2 * LumaBlue)Cr = ( R - Y ) / ( 2 - 2 * LumaRed)

where:LumaRed= 299/1000LumaGreen= 587/1000LumaBlue= 114/1000

Note: 64-bit arithmetic is the recommended precision for this conversion.

Since most of the perceivable informationcomes from the luminance, we can sample thechrominance half as often as we sample theluminance with only a minor change in imagequality.

Here, the sampling ratio used 4:2:2, whichmeans that there is a 2:1 ratio betweenluminance to each chrominance channel.


The images is divided below into 8x8 pixel blocks beforebeing encoded in the following MCU data order.

2. Partition into Blocks1.2. Partition into 8x8 Blocks

24-bit RGB (Original)

8-bit Luminance (Y) 8-bit Chrominance Blue (Cb) 8-bit Chrominance Red (Cr)


Chrominance Red (Cr)MCU0 =

…

Chrominance Blue (Cb)

= 8x8 Pixel BlockNote: For an MMX Image M = 40, N = 31

1.3. Minimum C oded U nits Data Ordering

0

1

…

Luminance (Y)0 1 … M - 1

0

1

…

N - 1

0 1 … M - 1

N - 1

0

1

…

N - 1

a b c

d

0 1 … M - 1

MCU1 =

e f g

h

1.3. Form M inimum C oded U nits

a b c d

e gf h


voidforward_DCT (){

double temp[N][N];double temp1 = 0;

for (int j = 0; j < N; j++) {C[0][j] = 1.0 / sqrt(N);Ct[j][0] = C[0][j];

}for (int i = 1; i < N; i++) {

for (int j = 0; j < N; j++) {C[i][j] = sqrt( 2. 0 / N ) *

cos( ( 2 * j + 1 ) * i * pi/ ( 2.0 * N ) ) ;

Ct[j][i] = C[i][j];}

for (int i = 0; i < N; i++ ) {for (int j = 0; j < N; j++) {

temp[i][j] = 0.0;for (int k = 0; k < N; k++)

temp[i][j] += ( Pixels[i][k] - 128 ) *Ct[k][j];

}}

Matrix-MultiplyShifted Pixels by Ct

Shift Pixel (source) between(-128) and 127

}

} Prepare a Cosine matrix, Ct, that is a linearcombination of the basis functions for the

transformation

Ct:0.353 0.49 0.461 0.415 0.353 0.277 0.191 0.0970.353 0.415 0.191 -0.097 -0.353 -0.49 -0.461 -0.2770.353 0.277 -0.191 -0.49 -0.353 0.097 0.461 0.4150.353 0.097 -0.461 -0.277 0.353 0.415 -0.191 -0.490.353 -0.097 -0.461 0.277 0.353 -0.415 -0.191 0.490.353 -0.277 -0.191 0.49 -0.353 -0.097 0.461 -0.4150.353 -0.415 0.191 0.097 -0.353 0.49 -0.461 0.2770.353 -0.49 0.461 -0.415 0.353 -0.277 0.191 -0.097

2.2. DCT by Matrix Multiplication2.2. DCT by Matrix Multiplication


35 - Lower Left 63 - Lower Right

0 - Upper Left 28 - Upper Right

2-D Representations of the Basis Coveragefrom the DCT Corner Values

IncreasingHorizontalFrequency

IncreasingVertical

Frequency

The position of the values in the matrixafter the DCT relates to their spatialfrequencies.

DC

2.3. DCT Result Ranked by Frequency Perception

0 1 5 6 14 15 27 28

2 4 7 13 16 26 29 42

3 8 12 17 25 30 41 43

9 11 18 24 31 40 44 53

10 19 23 32 39 45 52 54

20 22 33 38 46 51 55 60

21 34 37 47 50 56 59 61

35 36 48 49 57 58 62 63

2.3. DCT Result Ranked by Frequency Perception


The human eye functions like a lowpass filter, and so some of the high frequency informationcan be removed without perceiving any loss.


IncreasingVertical

Frequency

Integer rounding is used for division. Q values (8 bits) increase with spatial frequency.

3.1. Divide by Q to Filter Out High Frequencies

16 11 10 16 24 40 51 61

12 12 14 19 26 58 60 55

14 13 16 24 40 57 69 56

14 17 22 29 51 87 80 62

18 22 37 56 68 109 103 77

24 35 55 64 81 104 113 92

49 64 78 87 103 121 120 101

72 92 95 98 112 100 103 99


17 18 24 47 99 99 99 99

18 21 26 66 99 99 99 99

24 26 56 99 99 99 99 99

47 66 99 99 99 99 99 99

99 99 99 99 99 99 99 99

99 99 99 99 99 99 99 99

99 99 99 99 99 99 99 99

99 99 99 99 99 99 99 99

IncreasingVertical

Frequency

Luminance Q Table Chrominance Q Table

3.1. Divide by Q to Filter Out High Frequencies


FDCT Quantizer JPEG entropy encoder

Table SpecificationsTable Specifications

SourceImageData

4.0. Entropy Encoding4. Entropy Encoding


This pattern begins by mapping thethe value at the lowest spatialfrequency, the DC coefficient.

The first AC coefficient is next.

Now the Zig-Zag pattern

The last values to be coded are thosethat relate to the highest spatialfrequencies.

4.1. Follow Zig-Zag Coding Sequence4.1. Follow Zig-Zag Coding Sequence

0 1 5 6 14 15 27 28

2 4 7 13 16 26 29 42

3 8 12 17 25 30 41 43

9 11 18 24 31 40 44 53

10 19 23 32 39 45 52 54

20 22 33 38 46 51 55 60

21 34 37 47 50 56 59 61

35 36 48 49 57 58 62 63


Incr

ea

sing

Ve

rtic

alF

req

uenc

y


After the DCT followed by quantization, theresult might look like the matrix shown to theleft.

The first value to be encoded in a block is theDC coefficient. This is the most importantvalue of the block, so it can have up to 12 bitssigned precision. It is coded as the differencefrom the DC value of the previous block.

4.1. Follow Zig-Zag Coding Sequence4.2. Encoding the DC coefficient

0

23 1 0 2 0 1 0 0

-8 0 0 0 1 0 1 0

5 1 -2 0 0 0 0 0

0 -1 0 0 -1 0 0 -1

0 0 0 0 0 0 0 0

0 0 0 0 0 0 0

0 1 1 0 0 0 0 0

0 0 0 0 0 0 0 0


Incr

ea

sing

Ve

rtic

alF

req

uenc

y


Huffman entropy coding is used to describe thenumber of bits used for the integer. Thiscodeword precedes the integer and is called theVLC . The Huffman codes are kept in a tablesimilar to the one shown left.

4.1. Follow Zig-Zag Coding Sequence4.2.1 Encoding the DC VLC


Sz Co_len Co_wrd0 2 01 3 22 3 33 3 44 3 55 3 66 4 e7 5 1e8 6 3e9 7 7eA 8 feB 9 1fe


The Huffman table for theVLC describes therun-length pairs: the zero-run and the numberof zeros in the integer.

4.1. Follow Zig-Zag Coding Sequence4.3.1 Encoding the AC VLC


Run/Sz Co_len Co_wrd0/0 4 a0/1 2 00/2 2 10/3 3 4

…F/6 16 fffaF/7 16 fffbF/8 16 fffcF/9 16 fffdF/A 16 fffe


TheVLIs are encoded using the same representation. However, they range only from 0-10 bits.

4.1. Follow Zig-Zag Coding Sequence4.3.2 Encoding the AC coefficients

Integer Value-1,1

-3,-2, 2, 3-7..-4, 4..7

-15..-8, 8..15-31..-16, 16..31-63..-32, 32..63

-127..-64, 64..127-255..-128, 128..255-511..-256, 256..511

-1023..-512, 512..1023


10


00, 01, 10, 11000..011, 100..111

0000..0111, 1000..111100000..01111, 10000..11111

000000..011111, 100000..1111110000000..0111111, 1000000..1111111

00000000..01111111, 10000000..11111111000000000..011111111, 100000000..111111111

0000000000..0111111111, 1000000000..1111111111


23

Next, theAC coeffientsare encoded inzig-zag order.

Only thenon-zerocoeffients are encoded, sotheir position is kept track of by recording therun-length of zeros in between.

4.1. Follow Zig-Zag Coding Sequence4.3. Encoding the AC coefficients

0

0 0 0 0

0 0 0 0 0

0 0 0 0 0

0 0 0 0 0

0 0 0 0 0 0 0 0

0 0 0 0 0 0 0

0 0 0 0 0 0

0 0 0 0 0 0 0 0


Incr

ea

sing

Ve

rtic

alF

req

uenc

y

11

1

1 1

-1-1-1

-215

-8

21


The integer value, called theVLI , is coded as shown in this table. They range from 0-11 bits.

4.1. Follow Zig-Zag Coding Sequence4.2.2 Encoding the DC VLI

Integer Value-1,1

-3,-2, 2, 3-7..-4, 4..7

-15..-8, 8..15-31..-16, 16..31-63..-32, 32..63

-127..-64, 64..127-255..-128, 128..255-511..-256, 256..511

-1023..-512, 512..1023-2047..-1024, 1024..2047


1011


00, 01, 10, 11000..011, 100..111

0000..0111, 1000..111100000..01111, 10000..11111

000000..011111, 100000..1111110000000..0111111, 1000000..1111111

00000000..01111111, 10000000..11111111000000000..011111111, 100000000..111111111

0000000000..0111111111, 1000000000..111111111100000000000..01111111111, 10000000000..11111111111


4.0. Entropy Encoding

IDCTJPEG entropy decoder

Table SpecificationsTable Specifications

CompressedImageData

Dequantizer

UncompressedImageData

JPEG “Thumbnail” Decoder

“Thumbnail” Size

DCValues


4.0. Entropy EncodingJPEG Decoding Example – 0xFF Marker Codes

3EC1FFE0

Entropy-coded data:

00 DC55

SOI Marker Code

4E1FF923 00FF FF


4.0. Entropy EncodingJPEG Decoding Example – Serial Bit Shifts

Entropy-coded data:

0011111011000001111111110100111000011111001001

111110001111111111110111000101010

VLC

DC value (Bits = 0)

AC value (Zero Run = 1, Bits = 4)AC value (Zero Run = 6, Bits = 4)

AC value (Zero Run = 10, Bits = 4)

VLI

0-15-15

-14

AC value (Zero Run = 14, Bits = 4) -13

EOB

SOI Marker Code

jpeg processing modules for real-time decoding and …field programmable port extender (fpx) 6...

Documents