Lecture Slides on Advanced Video Coding | ECE 6258, Study notes of Digital Signal Processing

Material Type: Notes; Class: Digital Image Processing; Subject: Electrical & Computer Engr; University: Georgia Institute of Technology-Main Campus; Term: Fall 2003;

Typology: Study notes

Pre 2010

Uploaded on 08/05/2009

koofers-user-c7u
koofers-user-c7u 🇺🇸

5

(1)

10 documents

1 / 6

Toggle sidebar

This page cannot be seen from the preview

Don't miss anything!

bg1
1
11/17/2003 ECE 6258 Russell M. Mersereau 1
ECE6258 Lecture 35
Advanced Video Coding
11/17/2003 ECE 6258 Russell M. Mersereau 2
What is fixed and what is free
All of the image and video compression
standards define the syntax of the bit-stream
and the operation of the decoder.
The detailed operation of the encoder and
operation of some parts of the decoder are
left up to the manufacturer.
What can be optimized?
11/17/2003 ECE 6258 Russell M. Mersereau 3
Decoder options
Post-processing
Blocking artifact removal
There are many possible reconstructed images that are consistent
with the bit stream. Find the one that is the smoothest.
Error concealment
If a macroblock or a slice is lost in transmission, try to estimate it
using neighboring blocks, or blocks borrowed from earlier frames.
Image Enhancement
11/17/2003 ECE 6258 Russell M. Mersereau 4
Encoder Options
Distribution of I, P, and B frames
I-frames require the most bits, B-frames the
fewest
Method for motion estimation
Accuracy versus bit-rate versus complexity
Prediction mode for individual macroblocks
for P and B frames
Intracoded, forward prediction, backward
prediction, bidirectional prediction
1 or 4 motion vectors per macroblock
Slice size
pf3
pf4
pf5

Partial preview of the text

Download Lecture Slides on Advanced Video Coding | ECE 6258 and more Study notes Digital Signal Processing in PDF only on Docsity!

11/17/

ECE 6258 Russell M. Mersereau

1

ECE6258 Lecture 35

Advanced Video Coding

11/17/

ECE 6258 Russell M. Mersereau

What is fixed and what is free „^

All of the image and video compressionstandards define the syntax of the bit-streamand the operation of the decoder.

„^

The detailed operation of the encoder andoperation of some parts of the decoder areleft up to the manufacturer.

„^

What can be optimized?

11/17/

ECE 6258 Russell M. Mersereau

3

Decoder options „^

Post-processing ‰^

Blocking artifact removal

There are many possible reconstructed images that are consistentwith the bit stream. Find the one that is the smoothest. ‰^

Error concealment

If a macroblock or a slice is lost in transmission, try to estimate itusing neighboring blocks, or blocks borrowed from earlier frames. ‰^

Image Enhancement

11/17/

ECE 6258 Russell M. Mersereau

Encoder Options „^

Distribution of I, P, and B frames

I-frames require the most bits, B-frames thefewest

„^

Method for motion estimation

Accuracy versus bit-rate versus complexity

„^

Prediction mode for individual macroblocksfor P and B frames ‰^

Intracoded, forward prediction, backwardprediction, bidirectional prediction ‰^

1 or 4 motion vectors per macroblock

„^

Slice size

11/17/

ECE 6258 Russell M. Mersereau

5

Encoder Options (2) „^

Rate-Control ‰^

Needed to prevent buffer over/underflow ‰^

Allocation of bits to I, P, and B frames ‰^

Adjustment of quantization step size.

„^

Selection of quantization table

„^

Error resilience features (MPEG4) ‰^

Reversible Huffman codes ‰^

Resynchronization markers ‰^

Error-control codes ‰^

Motion vectors on I-frames

11/17/

ECE 6258 Russell M. Mersereau

ITU H.264 „^

This is the newest, and most powerful of thecoding standards (2003). ‰^

~2x coding gain over MPEG-

„^

Optimal implementations still under study.

„^

Other names for the same standard: ‰^

MPEG-4 Part 10 ‰^

MPEG-4 Advanced Video Coding (AVC) ‰^

H.26L ‰^

JVT

11/17/

ECE 6258 Russell M. Mersereau

7

Why people are excited

11/17/

ECE 6258 Russell M. Mersereau

Encoder Block Diagram

Deblocking

11/17/

ECE 6258 Russell M. Mersereau

13

4x4 Block-based Spatial Prediction „^

9 prediction modes(directions) ‰^

Mode 0 (DC) „^

Predict all pixels from(A+B+C+D+I+J+K+L)/ ‰^

Mode 1 (vertical) „^

a,e,I,m predicted from A;b,f,j,n from B … ‰^

Mode 2 (horizontal) „^

a,b,c,d from I; e,f,g,h fromJ, … ‰^

Mode 3 (NW-SE) „^

A,f,k,p from (A+2Q+I+2)/4; „^

e,j,o from (Q+2I+J+2)/4… ‰^

‰^

Mode 8

p o n m L

l k j i K

h g f e J

d c b a I

D C B A Q

11/17/

ECE 6258 Russell M. Mersereau

8x8 Spatial Prediction (Chrominance)^ „^

8 x 8 chroma macroblock consists offour 4 x 4 blocks ABCD.

„^

4 prediction cases depending uponwhether S0, S1, S2, or S3 are inside oroutside

„^

If all inside ‰^

A=(S0+S2+4)/

‰^

B=(S1+2)/

‰^

C=(S3+2)/

‰^

D=(S1+S3+4)/8S0, S1, S2, and S3 are the sums offour neighboring samples

A^

B

C

D

SO

S

S2 S

11/17/

ECE 6258 Russell M. Mersereau

15

Temporal prediction: multiple reference frames

Currentframe

11/17/

ECE 6258 Russell M. Mersereau

Motion Estimation „^

16 x 16 Macroblock ‰^

Prediction block structure „^

16x16, 16x8, 8x16, 8x8, 8x4, 4x8, 4x

0 0

0 0

0 0

(^0101)

1 1

1 1 2 2

3 3

MB-Modes 8x8-Modes

16x

16x

8x

8x

8x

8x

4x

4x

11/17/

ECE 6258 Russell M. Mersereau

17

Motion Vector Coding „^

Median prediction(except for 16x8 and8x16 modes)

„^

Median of A,B, and C(normally)

„^

A,B,C, and D can befrom different referenceframes

„^

¼ pixel resolution

D^

B

A

C

E

11/17/

ECE 6258 Russell M. Mersereau

B-Frame coding: motion vector coding „^

Modes ‰^

Direct, forward, backward, bi-directional, intra ‰^

16x16, 8x16, 16x8, 8x8, 4x8, 8x4, 4x

„^

Direct Mode ‰^

No MV data is transmitted ‰^

Same block structure as co-located MB intemporally subsequent picture ‰^

MVs are computed as scaled version ofcorresponding MV of the co-located block

11/17/

ECE 6258 Russell M. Mersereau

19

“DCT” Transform Computation (1) „^

Start with the 4x4 DCT

c

a

b

a

b

a

c

a

b

a

c

a

c a b a x x x x

x

x

x

x

x

x

x

x

x x x x c b b c

a

a

a

a

b

c

c

b

a

a

a

a

AXA

Y^

T

44 43 42 41

34 33 32 31

24 23 22 21

14 13 12 (^11) (

) (^

)^

... 65

(^3826834323). 0 (^8) / 3 cos 2 / 1

... 8

(^6532814243). 0 (^8) /

cos 2 (^2) / / 1 1

=

=

=

= =

π π

a b c

11/17/

ECE 6258 Russell M. Mersereau

“DCT” Transform Computation (2)^ „^

Manipulate

 





− −

− −







− −

− −



= =

b a b a d

d d

d

x x x x

x x x x

x x x x

x x x x d d

d d b a b a

B BCXC Y^

T

0 0 0

0 0 0

0 0 0

0 0 0 1 1 1

1 1

1

1 1

1

1 1 1

1 1

1 1 1 1

1

1

1 1 1 1 0 0 0

0 0 0

0 0 0

0 0 0

44 43 42 41

34 33 32 31

24 23 22 21

14 13 12 11

b c d^

/

(^

)

2

2

2

2

2

2

2

2

44 43 42 41

34 33 32 31

24 23 22 21

14 13 12 11

b ab b ab

ab a ab a

b ab b ab

ab a ab a d

d d

d

x x x x

x x x x

x x x x

x x x x d d

d d

E

CXC

Y^

T