



Study with the several resources on Docsity
Earn points by helping other students or get them with a premium plan
Prepare for your exams
Study with the several resources on Docsity
Earn points to download
Earn points by helping other students or get them with a premium plan
Material Type: Notes; Class: Digital Image Processing; Subject: Electrical & Computer Engr; University: Georgia Institute of Technology-Main Campus; Term: Fall 2003;
Typology: Study notes
1 / 6
This page cannot be seen from the preview
Don't miss anything!




11/17/
ECE 6258 Russell M. Mersereau
1
11/17/
ECE 6258 Russell M. Mersereau
11/17/
ECE 6258 Russell M. Mersereau
3
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
I-frames require the most bits, B-frames thefewest
Accuracy versus bit-rate versus complexity
Intracoded, forward prediction, backwardprediction, bidirectional prediction ^
1 or 4 motion vectors per macroblock
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) ^
Predict all pixels from(A+B+C+D+I+J+K+L)/ ^
a,e,I,m predicted from A;b,f,j,n from B … ^
a,b,c,d from I; e,f,g,h fromJ, … ^
A,f,k,p from (A+2Q+I+2)/4; ^
e,j,o from (Q+2I+J+2)/4… ^
^
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)^ ^
^
^
^
^
^
A^
B
C
D
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 ^
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
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
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)^ ^
−
−
− −
− −
−
−
− −
− −
= =
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
T