VIDEO CODECS

Director, eSILICON LABS, INDIA
VIDEO CODEC
Vinayagam M
Next Generation Broadcasting Technology

3
Agenda
HVS
Images / Video
Video / Image Compression
Image Coding
Video Coding
Video Coder Architecture
Video Codec Standards
HEVC

5
HVS
• HVS properties influence the
design/tradeoffs of imaging/video
systems
• Basic properties of HVS “front-end”
– 4 types of photo-receptors in the retina
– Rods, 3 types of cones
• Rods
– Achromatic (no concept of color)
– Used for scotopic vision (low light levels)
– Concentrated in periphery
• Cones
– 3 types: S - Short, M- Medium, L - Long
– Red, Green, and Blue peaks
– Used for Photopic Vision (daylight levels)
– Concentrated in fovea (center of the
retina)

6
HVS…
• Eyes, optic nerve, parts of the brain
• Transforms electromagnetic energy
• Image Formation
– Cornea, Sclera, Pupil, Iris, Lens, Retina, Fovea
• Transduction
– Retina, Rods, and Cones
• Processing
– Optic Nerve, Brain
• Retina and Fovea
– Retina has photosensitive receptors at back of eye
– Fovea is small, dense region of receptors
Only cones (no rods)
Gives visual acuity
– Outside Fovea
Fewer receptors overall
Larger proportion of rods
Fov
ea
Retina

7
HVS…
• Transduction (Retina)
– Transform light to neural impulses
– Receptors signal bipolar cells
– Bipolar cells signal ganglion cells
– Axons in the ganglion cells form optic
nerve
• Image Formation in the Human Eye

8
HVS…
• HVS Properties
– Tradeoff in resolution between space and time
Low resolution for high spatial AND high temporal frequencies
However, eye tracking can convert fast-moving object into low retinal frequency
– Achromatic versus chromatic channels
Achromatic channel has highest spatial resolution
Yellow/Blue has lower spatial resolution than Red/Green channel
– Color refers to how we perceive a narrow band of electromagnetic energy
Source, Object, Observer

9
HVS…
• Visual System
– Visual system transforms light energy into sensory experience of sight

10
HVS…
• Color Perception (Color Theory)
– Hue
Distinguishes named colors, e.g., RGB
Dominant wavelength of the light
– Saturation
Perceived intensity of a specific color
How far color is from a gray of equal intensity
– Brightness (lightness)
Perceived intensity
Hue Scale
SaturationOriginallightness

11
HVS…
• Visual Perception
– Resolution and Brightness
– Spatial Resolution depends on
Image Size
Viewing Distance
– Brightness
Perception of brightness is higher than perception of color
Different perception of primary colors
Relative Brightness: green:red:blue=59%:30%:11%
– B/W vs. Color

12
HVS…
• Visual Perception
– Temporal Resolution
Effects caused by inertia of human eye
Perception of 16 frames/second as continuous sequence
Special Effect: Flicker
Flicker
Perceived if frame rate or refresh rate of screen too low (<50Hz)
Especially in large bright areas
Higher refresh rate requires
Higher scanning frequency
Higher bandwidth

13
HVS…
• Visual Perception Influence
– Viewing distance
– Display ratio (width/height – 4/3 for conventional TV)
– Number of details still visible
– Intensity (luminance)

14
HVS…
• Imaging / Visual System designed
based on HVS principles
• Example
– Image Sensor
– Television
– Image / Video Display
• Image Sensor
– CCD (charge coupled device):
Arrays of photo diodes
Linearity
Less light needed
Electronic shuttering
– CMOS
Cheaper
Easy manufacturing
• Television
– NTSC (National Television System
Committee):
60 Hz, 30 fps, 525 scan lines
North America, Japan, Korea ….
– PAL (Phase Alteration by Line):
50 Hz, 25 fps, 625 scan lines
Europe …
• Image / Video Display
– CRT Monitor
– LCD TV/Display Monitor

16
IMAGE / VIDEO
• Images
– View Observation by HVS @ time instant
– A multidimensional array of numbers (such as intensity image) or vectors
(such as color image)
Each component in the image called pixel
associates with the pixel value (a single
number in the case of intensity images or
a vector in the case of color images)












39871532
22132515
372669
28161010












39656554
42475421
67965432
43567065












99876532
92438585
67969060
78567099

17
IMAGE / VIDEO…
• Video
– Series of Frames (or Images)

18
IMAGE / VIDEO…
• Images / Video Frame
– A multidimensional function of spatial coordinates
– Spatial Coordinate
(x,y) for 2D case such as photograph,
(x,y,z) for 3D case such as CT scan images
(x,y,t) for movies
– The function f may represent intensity (for monochrome images) or color
(for color images) or other associated values
Image “After snow storm” f(x,y)
x
y
Origin

19
IMAGE / VIDEO…
• Images / Video Frame
– An image that has been discretized both in Spatial coordinates and
associated value
Consist of 2 sets:(1) a point set and (2) a value set
Can be represented in the form
– I = {(x,a(x)): x ε X, a(x) ε F}
where X and F are a point set and value set, respectively
An element of the image, (x,a(x)) is called a pixel
where
x is called the pixel location and
a(x) is the pixel value at the location x
– Conventional Coordinate for Image Representation

20
IMAGE / VIDEO…
• Images / Video Frame Representation
– Basic Unit : Pixel
– Dimensions
Height
Width
– Frame rate determines how long the pixel
exists, i.e. how it moves
– Color Depth of the pixel
How many bits are used to represent the color of
each pixel?

21
IMAGE / VIDEO…
• Image Type
– Binary Image
– Intensity Image
– Color Image
– Index image

22
IMAGE / VIDEO…
• Binary Image
– Binary image or black and white image
– Each pixel contains one bit
1 represent white
0 represents black












1111
1111
0000
0000
Binary Data

23
IMAGE / VIDEO…
• Intensity Image
– Intensity / Monochrome/ Gray Scale Image
– Each pixel corresponds to light intensity normally represented in gray
scale (gray level)












39871532
22132515
372669
28161010
Gray Scale Values

24
IMAGE / VIDEO…
• Color Image
– Each pixel contains a vector representing red, green and blue components












39871532
22132515
372669
28161010












39656554
42475421
67965432
43567065












99876532
92438585
67969060
78567099
RGB Components

25
IMAGE / VIDEO…
• Index Image
– Each pixel contains index number pointing to a color in a color table










256
746
941
Index Value
Index
No.
Red
component
Green
component
Blue
component
1 0.1 0.5 0.3
2 1.0 0.0 0.0
3 0.0 1.0 0.0
4 0.5 0.5 0.5
5 0.2 0.8 0.9
… … … …
Color Table

26
IMAGE / VIDEO…
• Colourspace Representations
– RGB (Red, Green, Blue) – Basic analog components (from camera/to TV)
– YPbPr (Y,B-Y,R-Y) – ANALOG Colourspace (derived from RGB)
Y=Luminance, B=Blue,
– R=Red
– YUV – Colour difference signals scaled to be modulated on a composite
carrier
– YIQ – Used in NTSC. I=In-phase, Q=Quadrature (IQ plane is 33o rotation
of UV plane)
– YCbCr/YCC – DIGITAL representation of the YPbPr Colourspace (8bit, 2s
compliment)

27
IMAGE / VIDEO…
• RGB Color
– All color can be composed by adding specific amounts of R, G, & B
– 8-bits (28) specifies the amount of each color
– This is the scheme used by most electronic displays to generate color;
e.g. we often call our computer monitors, "RGB displays"
8-bits Red
8-bits Green
8-bits Blue

28
IMAGE / VIDEO…
• Color Reduction
– Human eye is not as sensitive to color as it is to Luminance
– To this end, to save costs the various standards decided to
Maintain luminance information in our images, but Reduce color information
Using RBG, though, how do we easily reduce color information without
removing luminance?
For this, and other technical reasons, a separate color space was chosen by
most video standards …

29
IMAGE / VIDEO…
• Colour Image: RGB
• YCbCr
– Even though most displays actually
use RGB to create the image, YCbCr
is used most often in consumer
electronics for transmission of the
image
– Historically, B/W televisions
transmitted only luminance (Y)
– The color signals were added later

30
IMAGE / VIDEO…
• YCbCr Generated By Sub sampling
– YUV 4:4:4 = 8bits per Y,U,V channel (no downsampling the chroma
channels)
– YUV 4:2:2 = 4 Y pixels sampled for every 2 U and 2 V (2:1 horizontal
downsampling, no vertical downsampling
– YUV 4:2:0 = 2:1 horizontal downsampling, 2:1 vertical downsampling
– YUV 4:1:1 = 4 Y pixels sampled for every 1 U and 1 V (4:1 horizontal
downsampling, no vertical downsampling)
• YUV 4:4:4
Y Y Y Y
Y Y Y Y
4:4:4 Format (3 bytes/pixel):
Cb Cr Cb Cr Cb Cr Cb Cr
Cb Cr Cb Cr Cb Cr Cb Cr

31
IMAGE / VIDEO…
• YUV 4:2:2
• YUV 4:2:0
Y Y Y Y
Y Y Y Y
4:2:2 Format (2 bytes/pixel):
Cb Cr
Cb Cr
Cb Cr
Cb Cr
Y Y Y Y
Y Y Y Y
Cb Cr Cb Cr
4:2:0 Format (1.5 bytes/pixel):

32
IMAGE / VIDEO…
• Up sampling
• Downsampling
nT
Input Signal
1 2 3 4
F(nT) F(nT/2)
nT
Intermediate Signal
12345678
Interpolating
low-pass filter
nT
nT
F(nT/2)
Output Signal
12345678
nT
Input Signal
1 2 3 4
F(nT)
Decimating
low-pass filter
prevents alias
at lower rate
F(2nT)
1
Output Signal
2

33
IMAGE / VIDEO…
• RGB to YCbCr
• RGB to YUV Conversion
– Y = 0.299R + 0.587G + 0.114B
– U= (B-Y)*0.565
– V= (R-Y)*0.713
U-V plane at Y=0.5
Clamp the output: Y=[16, 235], U,V=[16,239]

35
VIDEO/IMAGE COMPRESSION
• How can we use fewer bits?
• To understand how image/audio/video signals are compressed to
save storage and increase transmission efficiency
• Reduces signal size by taking advantage of correlation
– Spatial
– Temporal
– Spectral

36
VIDEO/IMAGE COMPRESSION…
• Compression Methods
• Need to take advantage of redundancy
– Images
Space
Frequency
– Video
Space
Frequency
Time
Linear Predictive AutoRegressive Polynomial Fitting
Model-Based
Huffman
Statistical
Arithmetic Lempel-Ziv
Universal
Lossless
Spatial/Time-Domain
Subband Wavelet
Filter-Based
Fourier DCT
Transform-Based
Frequency-Domain
Lossy
Waveform-Based
Compression Methods

37
• Need to take advantage of redundancy
RGB
YCbCr
Blocks
Macro
Blocks
I B P
Remove Temporal Redundancy
Transform
QuantizationCoding
01100010101,0
Remove Spatial Redundancy
Motion
Compensation

38
• Spatial Redundancy
– Take advantage of similarity among most neighboring
pixels
• RGB to YUV
– Less information required for YUV (humans less sensitive
to chrominance)
• Macro Blocks
– Take groups of pixels (16x16)
• Discrete Cosine Transformation (DCT)
– Based on Fourier analysis where represent signal as sum
of sine's and cosine’s
– Concentrates on higher-frequency values
– Represent pixels in blocks with fewer numbers
• Quantization
– Reduce data required for coefficients
• Entropy coding
– Compress

39
• Spatial Redundancy Reduction
Zig-Zag Scan,
Run-length
coding
Quantization
• major reduction
• controls ‘quality’
“Intra-Frame
Encoded”

40
• When may spatial redundancy elimination be ineffective?
– High-resolution images and displays
– May appear ‘coarse’
• What kinds of images/movies?
– A varied image or ‘busy’ scene
– Many colors, few adjacent
Original (63 kb)
Low (7kb)
Very Low (4 kb)Due to Loss of Resolution
Solution ? Temporal Redundancy Reduction

41
• Temporal Redundancy Reduction
– Take advantage of similarity between successive frames
950 951 952

42

43

44
When may temporal redundancy
reduction be ineffective?

45
• Many scene changes vs. few scene changes
• Sometimes high motion

46
• Many scene changes vs. few scene changes
• Sometimes high motion

48
IMAGE CODING
• Lossless Compression
• Lossy Compression
• Transform Coding

49
IMAGE CODING…
• Image compression system is composed of three key building blocks
– Representation
Concentrates important information into a few parameters
– Quantization
Discretizes parameters
– Binary encoding
Exploits non-uniform statistics of quantized parameters
Creates bitstream for transmission

50
IMAGE CODING…
• Image compression system is composed of three key building blocks
– Representation
Concentrates important information into a few parameters
– Quantization
Discretizes parameters
– Binary encoding
Exploits non-uniform statistics of quantized parameters
Creates bitstream for transmission

51
IMAGE CODING…
• Generally, the only operation that is lossy is the quantization stage
• The fact that all the loss (distortion) is localized to a single operation
greatly simplifies system design
• Can design loss to exploit human visual system (HVS) properties
• Source decoder performs the inverse of each of the three operations

52
IMAGE CODING…
• Representations - Transform and Subband Filtering Methods
– Goal
Transform signal into another domain where most of the information (energy) is
concentrated into only a small fraction of the coefficients
– Enables perceptual processing
Exploiting HVS response to different frequency components

53
IMAGE CODING…
• Representations - Transform and Subband Filtering Methods
– Examples of “traditional” transforms
KLT, DFT, DCT
– Examples of “traditional” Subband filtering methods
Perfect reconstruction filter banks, wavelets
– Transform and Subband interpretations
All of the above are linear representations and can be interpreted from either a
transform or a Subband filtering viewpoint
– Transform viewpoint
Express signal as a linear combination of basis vectors
Stresses linear expansion (linear algebra) perspective
– Subband filtering viewpoint
Pass signal through a set of filters and examine the frequencies passed by
each filter (Subband)
Stresses filtering (signal processing) perspective

54
IMAGE CODING…
• Representations – Transform Image Coding
– A good transform provides
Most of the image energy is concentrated into a small fraction of the
coefficients
Coding only these small fraction of the coefficients and discarding the rest can
often lead to excellent reconstructed quality
The more energy compaction the better
– Orthogonal transforms are particularly useful
Energy in discarded coefficients is equal to energy in reconstruction error

55
IMAGE CODING…
• Representations – Transform Image Coding
– Karhunen-Loeve Transform (KLT)
Optimal energy compaction
Requires knowledge of signal covariance
In general, no simple computational algorithm
– Discrete Fourier Transform (DFT)
Fast algorithms
Good energy compaction, but not as good as DCT
– Discrete Cosine Transform (DCT)
Fast algorithms
Good energy compaction
All real coefficients
Overall good performance and widely used for image and video coding

56
IMAGE CODING…
• Discrete Cosine Transform (DCT)
– 1-D Discrete Cosine Transform (N-point)
– 1-D DCT basis vectors
– 2-D DCT: Separable transform of 1-D DCT
– 2-D DCT basis vectors?
Basis pictures!
– 2-D basis vectors for 2-D DCT are basis pictures!
– 64 basis pictures for 8x8-pixel 2-D DCT
– Image coding with the 2-D DCT is equivalent to approximating the image
as a linear combination of these basis pictures!

57
IMAGE CODING…
• Representations – Coding Transform Coefficients
– Selecting the basis pictures to approximate an image is equivalent to
selecting the DCT coefficients to code
– General methods of coding/discarding coefficients
Zonal Coding
▫ Code all coefficients in a zone and discard others
▫ Example zone: Spatial low frequencies
▫ Only need to code coefficient amplitudes
Threshold Coding
▫ Keep coefficients with magnitude above a threshold
▫ Coefficient amplitudes and locations must be coded
▫ Provides best performance

58
IMAGE CODING…
• Video / Image Coding are Block
based Coding
– Frames are divided into Sub-Block
and then coded
• Macroblock (MB) and Block Layer
– Process the data in blocks of 8x8
samples
– Convert Red-Green-Blue into
Luminance (greyscale) and
Chrominance (Blue color difference
and Red color difference)
– Use half resolution for Chrominance
(because eye is more sensitive to
greyscale than to color)

59
IMAGE CODING…
• Macroblock (MB) and Block Layer
– Macroblock
Consist of
16x16 luminance block
8x8 chrominance block
Basic unit for motion estimation
– Block
8 pixels by 8 lines
Basic unit for DCT

60
IMAGE CODING…
– General-Purpose Compression: Entropy Encoding
– Remove statistical redundancy from data
– ie, Encode common values with short codes, uncommon values with
longer codes
– Huffman Coding
– Example : ABCCDEAAB
After compression: 1011000000001010111011
– Compression ratio
According to probability of the characters appears in the uncompressed
data
C:12 D:13 F:5 E:9 B:16 A:45
1425
55
100
30
10
0
0
0
0
1
1
1
1
000 001 0100 0101 011 1

61
IMAGE CODING…
– Run-Length Coding
Reduce the number of samples to code
Implementation is simple
Input Sequence
0,0,-3,5,1,0,-2,0,0,0,0,2,-4,3,-2,0,0,0,1,0,0,-2,EOB
Run-Length Sequence
(2,-3)(0,5)(0,1)(1,-2)(4,2)(0,-4)(0,3)(0,-2)(3,1)(2,-2)EOB

62
IMAGE CODING…
– Transform Coding
(-1,1) (1,1)
(0.4,1.4) = 0.4•(1,0)+1.4•(0,1)
= 0.9•(1,1)+0.5•(-1,1)
Basis vector
{ (1,0), (0,1) }
New basis vector
{ (1,1), (-1,1) }
New vector
(0.9, 0.5)
(0,1)
(1,0)

63
IMAGE CODING…
– Transform Coding : DCT
Transform blocks of images to frequency domain, code only the significant
transform coefficients
2D DCT
8x8 DCT Basis Function

64
IMAGE CODING…
2D DCT Coefficients

65
IMAGE CODING…
– Lossy Predictive Coding

66
IMAGE CODING…
– Quantization
Many to one mapping
Quantization is the most import means of irrelevancy reduction
– Implementation
Lookup Table
Divide by quantization step-size (round/truncate)

67
IMAGE CODING…
– Divide by quantization step-size
Input signal：0 1 2 3 4 5 6 7（3 bits）
Step-size：2
Quantization：0 0 1 1 2 2 3 3（2 bits）
Inverse quantization：0 0 2 2 4 4 6 6
Quantization Errors：0 1 0 1 0 1 0 1
– Lookup Table
Divide each DCT coefficient by an integer, discard remainder
Result: loss of precision
Typically, a few non-zero coefficients are left

68
IMAGE CODING…
– Zigzag Scan
Efficient encoding of the position
of non-zero transform
coefficients
Scan” quantized coefficients in a
zig-zag order
Non-zero coefficients tend to be
grouped together

69
IMAGE CODING…
• DCT + Quantization + Run-Level-Coding

71
VIDEO CODING
• Transform Coding
• Motion Coding

72
VIDEO CODING…
• Video
– Sequence of frames (images) that are related
• Moving images contain significant temporal redundancy
– Successive frames are very similar
– Related along the temporal dimension - Temporal redundancy exists

73
VIDEO CODING…
• Video Coding
– The objective of video coding is to compress moving images
– Main addition over image compression
Temporal redundancy
Video coder must exploit the temporal redundancy
– The MPEG (Moving Picture Experts Group) and H.26X are the major
standards for video coding
• Video coding algorithms usually contains two coding schemes :
– Intraframe coding
Intraframe coding does not exploit the correlation among adjacent
frames
Intraframe coding therefore is similar to the still image coding
– Interframe coding
The interframe coding should include motion estimation/compensation process
to remove temporal redundancy
• Basic Concept
– Use interframe correlation for attaining better rate distortion

74
VIDEO CODING…
• Usually high frame rate: Significant temporal redundancy
• Possible representations along temporal dimension
– Transform/Subband Methods
Good for textbook case of constant velocity uniform global motion
Inefficient for nonuniform motion, I.e. real-world motion
Requires large number of frame stores
Leads to delay (Memory cost may also be an issue)
– Predictive Methods
Good performance using only 2 frame stores
However, simple frame differencing in not enough

75
VIDEO CODING…
• Main addition over image compression
– Exploit the temporal redundancy
• Predict current frame based on previously coded frames
• Types of coded frames
– I-frame
Intra-coded frame, coded independently of all other frames
– P-frame
Predictively coded frame, coded based on previously coded frame
– B-frame
Bi-directionally predicted frame, coded based on both previous and future coded
frames

76
VIDEO CODING…
• Motion-Compensated Prediction
– Simple frame differencing fails when there is motion
– Must account for motion
Motion-compensated (MC) prediction
– MC-prediction generally provides significant improvements
– Questions
How can we estimate motion?
How can we form MC-prediction?
• Motion Estimation
– Ideal Situation
Partition video into moving objects
Describe object motion
Generally very difficult
– Practical approach: Block-Matching Motion Estimation
Partition each frame into blocks
Describe motion of each block
No object identification required
Good, robust performance

77
VIDEO CODING…
• Block-Matching Motion Estimation
– Assumptions
Translational motion within block
All pixels within each block have the same motion
– ME Algorithm
Divide current frame into non-overlapping N1xN2 blocks
For each block, find the best matching block in reference frame
– MC-Prediction Algorithm
Use best matching blocks of reference frame as prediction of blocks in current
frame

78
VIDEO CODING…
• Block-Matching - Determining the Best Matching Block
– For each block in the current frame search for best matching block in the
reference frame
Metrics for determining “best match”
Candidate blocks: All blocks in, e.g., (± 32,±32) pixel area
Strategies for searching candidate blocks for best match
Full search: Examine all candidate blocks
Partial (fast) search: Examine a carefully selected subset
– Estimate of motion for best matching block: “motion vector”
• Motion Vectors and Motion Vector Field
– Motion Vector
Expresses the relative horizontal and vertical offsets (mv1,mv2), or motion, of
a given block from one frame to another
Each block has its own motion vector
– Motion Vector Field
Collection of motion vectors for all the blocks in a frame

79
VIDEO CODING…
• Example of Fast Search: 3-Step
(Log) Search
– Goal: Reduce number of search
points
Example:(± 7,±7) search area
Dots represent search points
Search performed in 3 steps
(coarse-to-fine)
– Step 1: (± 4 pixels )
– Best match is found at each step
– Next step: Search is centered
around the best match of prior step
– Speedup increases for larger
search areas

80
VIDEO CODING…
• Motion Vector Precision
– Motivation
Motion is not limited to integer-pixel offsets
However, video only known at discrete pixel locations
To estimate sub-pixel motion, frames must be spatially interpolated
– Fractional MVs are used to represent the sub-pixel motion
– Improved performance (extra complexity is worthwhile)
– Half-pixel ME used in most standards: MPEG-1/2/4
– Why are half-pixel motion vectors better?
Can capture half-pixel motion
Averaging effect (from spatial interpolation) reduces prediction error ->
Improved prediction
For noisy sequences, averaging effect reduces noise -> Improved
compression

81
VIDEO CODING…
• Practical Half-Pixel Motion Estimation Algorithm
– Half-Pixel ME (coarse-fine) Algorithm
Coarse Step: Perform integer motion estimation on blocks; find best integer-
pixel MV
Fine Step: Refine estimate to find best half-pixel MV
Spatially interpolate the selected region in reference frame
Compare current block to interpolated reference frame block
Choose the integer or half-pixel offset that provides best match
Typically, bilinear interpolation is used for spatial interpolation
• Example
– MC-Prediction for Two Consecutive Frames

82
VIDEO CODING…
• Bi-Directional MC-Prediction
– Bi-Directional MC-Prediction is used to estimate a block in the current
frame from a block in
Previous frame
Future frame
Average of a block from the previous frame and a block from the future frame
– Motion compensated prediction
Predict the current frame based on reference frame(s) while compensating for
the motion
– Examples of block-based motion-compensated prediction (P-frame)
and bi-directional prediction (B-frame)

83
VIDEO CODING…
• Motion Estimation and Compensation
– The amount of data to be coded can be reduced significantly if the
previous frame is subtracted from the current frame

84
VIDEO CODING…
– Uses Block-Matching
The MPEG and H.26X standards use block-matching technique for motion
estimation /compensation
In the block-matching technique, each current frame is divide into equal-size
blocks, called source blocks
Each source block is associated with a search region in the reference frame
The objective of block-matching is to find a candidate block in the search
region best matched to the source block
The relative distances between a source block and its candidate blocks are
called motion vectors
Video Sequence
The current frameThe reconstructed
reference frame
Bx: Search area
associated with X
MV: Motion Vector
X: Source block for
block-matching

85
VIDEO CODING…
– Uses Block-Matching

86
VIDEO CODING…
The Reconstructed Previous Frame The Current Frame
Results of Block-
Matching
The Predicted
Current Frame

87
VIDEO CODING…
– Search Range
– The size of the search range =
– The number of candidate blocks =
)2)(2( max2max1 yx dNdN  
)12)(12( maxmax   yx dd

88
VIDEO CODING…
– Motion Vector and Search Area
   pnpn 22 Search Area:
Motion vector: (u, v)

89
VIDEO CODING…
– Matching Function
Mean square error(MSE)
Mean absolute difference(MAD)
Number of threshold difference(NTD)
Normalized cross-correlation function(NCF)






1
0
21
0
221121
21
21
1
1
1
1
)]1,,(),,([
1
),(
N
n
N
n
tdndnftnnf
NN
ddMSE






1
0
1
0
221121
21
21
1
1
1
1
|)1,,(),,(|
1
),(
N
n
N
n
tdndnftnnf
NN
ddMAD

90
VIDEO CODING…
– Algorithm
Full search block matching (FSB)
Fast algorithm
▫ 2D Logarithmic Search (TDL)
▫ Three Step Search (TSS)
▫ Cross-Search Algorithm (CSA)
▫ …
– Full Search Algorithm
If p=7, then there are
(2p+1)(2p+1)=225 candidate blocks.
u
v
Search Area
Candidate
Block

91
VIDEO CODING…
– Full Search Algorithm
Intensive computation
Need for fast Motion Estimation !

92
VIDEO CODING…
– 2D Logarithmic Search
Diamond-shape search area
Matching function
▫ MSE
-7 –6 –5 –4 –3 –2 –1 0 +1 +2 +3 +4 +5 +6 +7
+7
+6
+5
+4
+3
+2
+1
0
-1
-2
-3
-4
-5
-6
-7
MV

93
VIDEO CODING…
– Three-Step Search
The first step involves block-matching based on 4-pel resolution at the nine
location
The second step involves block-matching based on 2-pel resolution around the
location determined by the first step
The third step repeats the process in the second step (but with resolution 1-pel)
-6
-5
-4
-3
-2
-1
0
1
2
3
4
5
6
7
11 1
11
11 1
-7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7
3
3
333
3
3 3
2 2 2
2
222
2

94
VIDEO CODING…
– Motion Vector Prediction
predMVx = Median(MV1x, MV2x, MV3x)
predMVy = Median(MV1y, MV2y, MV3y)
MVx`=MVx - predMVx
MVy`=MVy - predMVy

96
VIDEO CODER ARCHITECTURE
• Image / Video Coding Based on Block-Matching
– Assume frame f-1 has been encoded and reconstructed, and frame f is the
current frame to be encoded
• Exploiting the redundancies
– Temporal
MC-Prediction (P and B frames)
– Spatial
Block DCT
– Color
Color Space Conversion
• Scalar quantization of DCT coefficients
• Zigzag scanning, runlength and Huffman coding of the nonzero
quantized DCT coefficients

97
VIDEO CODER ARCHITECTURE…
• Video Encoder
– Divide frame f into equal-size blocks
– For each source block,
Find its motion vector using the block-matching algorithm based on the
reconstructed frame f-1
Compute the DFD of the block
– Transmit the motion vector of each block to decoder
– Compress DFD’s of each block
– Transmit the encoded DFD’s to decoder

98
• Video Encoder

99
• Video Decoder
– Receive motion vector of each block from encoder
– Based on the motion vector ,find the best-matching block from the
reference frame
ie,, Find the predicted current frame from the reference frame
– Receive the encoded DFD of each block from encoder
– Decode the DFD.
– Each reconstructed block in the current frame = Its decompressed DFD +
the best-matching block

100
• Video Decoder

102
VIDEO CODEC STANDARDS
• Goal of Standards
– Ensuring Interoperability
Enabling communication between devices made by different manufacturers
– Promoting a technology or industry
– Reducing costs
What do the Standards Specify?

103
VIDEO CODEC STANDARDS…
What do the Standards Specify?
• Not the encoder
• Not the decoder
• Just the bitstream syntax and the decoding process(e.g. use IDCT,
but not how to implement the IDCT)
– Enables improved encoding & decoding strategies to be employed in a
standard-compatible manner

104
• The Scope of Picture and Video Coding Standardization
– Only the Syntax and Decoder are standardized:
Permits optimization beyond the obvious
Permits complexity reduction for implementability
Provides no guarantees of Quality
Pre-Processing Encoding
Source
Destination
Post-Processing
& Error Recovery
Decoding
Scope of Standard

106
• Based on the same fundamental building blocks
– Motion-compensated prediction (I, P, and B frames)
– 2-D Discrete Cosine Transform (DCT)
– Color space conversion
– Scalar quantization, runlengths, Huffman coding
• Additional tools added for different applications:
– Progressive or interlaced video
– Improved compression, error resilience, scalability, etc.
• MPEG-1/2/4, H.261/3/4
– Frame-based coding
• MPEG-4
– Object-based coding and Synthetic video

107
• The Video Standards uses all the three types of frames as shown
below
Encoding order: I0, P3, B1, B2, P6, B4, B5, I9, B7, B8.
Playback order: I0, B1, B2, P3, B4, B5, P6, B7, B8, I9.

108
• Video Structure
– Video standards code video sequences in hierarchy of layers
– There are usually 5 Layers
GOP (Group of Pictures)
Picture
Slice
Macroblock
Block

109
• Video Structure
– A GOP usually started with I frame, followed by a sequence of P and B
frames
– A Picture is indeed a frame in the video sequence
– A Slice is a portion in a picture
Some standards do not have slices
Some view a slice as a row
Each slice in H.264 is not necessary to be a row
It can be any shape containing integral number of macroblocks
– A Macroblock is a 16×16 block
Many standards use Marcoblocks as the basic unit for block-matching
operations
– A Block is a 8×8 block
Many standards use the Blocks as the basic unit for DCT

110
• Scalable Video Coding
– Three classes of scalable video coding techniques
Temporal Scalability
Spatial Scalability
SNR Scalability
– Uses B frames for attaining temporal scalability
B frames depend on other frames
No other frames depend on B frames
Discard B frames without affecting other frames

111
• Scalable Video Coding – Spatial Scalability
– Basically Resolution Scalability
Here the base layer is the low resolution version of the video sequence
– The base layer uses coaster quantizer for DFD coding
– The residuals in the base layer is refined in the enhancement layer

114
HEVC
• Video Coding Standards Overview
Next Generation Broadcasting

115
HEVC…
• MPEG-H
– High Efficiency Coding and Media Delivery in
Heterogeneous Environments a new suite of
standards providing technical solutions for
emerging challenges in multimedia industries
– Part 1: System, MPEG Media Transport (MMT)
Integrated services with multiple components in a hybrid
delivery environment, providing support for seamless and
efficient use of heterogeneous network environments,
including broadcast, multicast, storage media and mobile
networks
– Part 2: Video, High Efficiency Video Coding
(HEVC)
Highly immersive visual experiences, with ultra high definition
displays that give no perceptible pixel structure even if
viewed from such a short distance that they subtend a large
viewing angle (up to 55 degrees horizontally for 4Kx2K
resolution displays, up to 100 degrees for 8Kx4K)
– Part 3: Audio, 3D-Audio
Highly immersive audio experiences in which the decoding
device renders a 3D audio scene. This may be using 10.2 or
22.2 channel configurations or much more limited speaker
configurations or headphones, such as found in a personal
tablet or smartphone.

116
HEVC…
• Transport/System Layer Integration
– On going definitions (MPEG, IETF,…,DVB): benefit from H.264/AVC
– MPEG Media Transport (MMT) ?

117
HEVC…
• HEVC = High Efficiency Video Coding
• Joint project between ISO/IEC/MPEG and ITU-T/VCEG
– ISO/IEC: MPEG-H Part 2 (23008-2)
– ITU-T: H.265
• JCT-VC committee
– Joint Collaborative Team on Video Coding
– Co-chairs: Dr. Gary Sullivan (Microsoft, USA) and Dr. Jens-Reiner Ohm (RWTH
Aachen, Germany)
• Target
– Roughly half the bit-rate at the same subjective quality compared to H.264/AVC (50%
over H.264/AVC)
– x10 complexity max for encoder and x2/3 max for decoder
• Requirements
– Progressive required for all profiles and levels
Interlaced support using field SEI message
– Video resolution: sub QVGA to 8Kx4K, with more focus on higher resolution video
content (1080p and up)
– Color space and chroma sampling: YUV420, YUV422, YUV444, RGB444
– Bit-depth: 8-14 bits
– Parallel Processing Architecture

118
HEVC…
• H.264 Vs H.265

119
HEVC…
• Potential applications
– Existing applications and usage scenarios
IPTV over DSL : Large shift in IPTV eligibility
Facilitated deployment of OTT and multi-screen services
More customers on the same infrastructure: most IP traffic is video
More archiving facilities
– Existing applications and usage scenarios
1080p60/50 with bitrates comparable to 1080i
Immersive viewing experience: Ultra-HD (4K, 8K)
Premium services (sports, live music, live events,…): home theater, Bars
venue, mobile
HD 3DTV Full frame per view at today’s HD delivery rates
What becomes possible with 50% video rate reduction?

120
HEVC…
• Tentative Timeline

122
HEVC…
• H.264 Vs H.265

123
HEVC…
• H.264 Vs H.265

126
HEVC…
• Video Coding Techniques : Block-based hybrid video coding
– Interpicture prediction
Temporal statistical dependences
– Intrapicture prediction
Spatial statistical dependences
– Transform coding
Spatial statistical dependences
• Uses YCbCr color space with 4:2:0 subsampling
– Y component
Luminance (luma)
Represents brightness (gray level)
– Cb and Cr components
Chrominance (chroma).
Color difference from gray toward blue and red

127
HEVC…
• Video Coding Techniques : Block-based hybrid video coding
– Motion compensation
Quarter-sample precision is used for the MVs
7-tap or 8-tap filters are used for interpolation of fractional-sample
positions
– Intrapicture prediction
33 directional modes, planar (surface fitting), DC (flat)
Modes are encoded by deriving most probable modes (MPMs) based
on those of previously decoded neighboring PBs
– Quantization control
Uniform reconstruction quantization (URQ)
– Entropy coding
Context adaptive binary arithmetic coding (CABAC)
– In-Loop deblocking filtering
Similar to the one in H.264 and More friendly to parallel processing
– Sample adaptive offset (SAO)
Nonlinear amplitude mapping
For better reconstruction of amplitude by histogram analysis

128
HEVC…
• Coding Tree Unit (CTU) - A picture is partitioned into CTUs
– The CTU is the basic processing unit instead of Macro Blocks (MB)
– Contains luma CTBs and chroma CTBs
A luma CTB covers L × L samples
Two chroma CTBs cover each L/2 × L/2 samples
– HEVC supports variable-size CTBs
The value of L may be equal to 16, 32, or 64.
Selected according to needs of encoders - In terms of memory and
computational requirements
Large CTB is beneficial when encoding high-resolution video content
– CTBs can be used as CBs or can be partitioned into multiple CBs using
quadtree structures
– The quadtree splitting process can be iterated until the size for a luma
CB reaches a minimum allowed luma CB size (8 × 8 or larger).

129
HEVC…
• Block Structure
– Coding Tree Units (CTU)
Corresponds to macroblocks in earlier coding standards (H.264, MPEG2, etc)
Luma and chroma Coding Tree Blocks (CTB)
Quadtree structure to split into Coding Units (CUs)
16x16, 32x32, or 64x64, signaled in SPS

130
HEVC…
• A new framework composed of three
new concepts
– Coding Units (CU)
– Prediction Units (PU)
– Transform Units (TU)
• The decision whether to code a
picture area using inter or intra
prediction is made at the CU level
Goal: To be as flexible as possible and to adapt the
compression-prediction to image peculiarities

131
HEVC…
• Block Structure
– Coding Units (CU)
Luma and chroma Coding Blocks (CB)
Rooted in CTU
Intra or inter coding mode
Split into Prediction Units (PUs) and Transform Units (TUs)

132
HEVC…
• Block Structure
– Prediction Units (PU)
Luma and chroma Prediction Blocks (PB)
Rooted in CU
Partition and motion info

133
HEVC…
• Block Structure
– Transform Units (TU)
Rooted in CU
4x4, 8x8, 16x16, 32x32 DCT, and 4x4 DST

134
HEVC…
• Relationship of CU, PU and TU

135
HEVC…
• Intra Prediction
– 35 intra modes: 33 directional modes +
DC + planar
– For chroma, 5 intra modes: DC, planar,
vertical, horizontal, and luma derived
– Planar prediction (Intra_Planar)
Amplitude surface with a horizontal and
vertical slope derived from boundaries
– DC prediction (Intra_DC)
Flat surface with a value matching the
mean value of the boundary samples
– Directional prediction (Intra_Angular)
33 different directional prediction is
defined for square TB sizes from 4×4 up
to 32×32

136
HEVC…
– Adaptive reference sample filtering
3-tap filter: [1 2 1]/4
Not performed for 4x4 blocks
For larger than 4x4 blocks, adaptively performed for a subset of modes
Modes except vertical/near-vertical, horizontal/near-horizontal, and DC
– Mode dependent adaptive scanning
4x4 and 8x8 intra blocks only
All other blocks use only diagonal upright scan (left-most scan pattern)

137
HEVC…
– Boundary smoothing
Applied to DC, vertical, and horizontal modes, luma only
Reduces boundary discontinuity
– For DC mode, 1st column and row of samples in predicted block are
filtered
– For Hor/Ver mode, first column/row of pixels in predicted block are filtered

138
HEVC…
• Inter Prediction
– Fractional sample interpolation
¼ pixel precision for luma
– DCT based interpolation filters
8-/7- tap for luma
4-tap for chroma
Supports 16-bit implementation
with non-normative shift
– High precision interpolation and
biprediction
– DCT-IF design
Forward DCT, followed by
inverse DCT

139
HEVC…
• Inter Prediction
– Asymmetric Motion Partition (AMP) for Inter PU
– Merge
Derive motion (MV and ref pic) from spatial and
temporal neighbors
Which spatial/temporal neighbor is identified by
merge_idx
Number of merge candidates (≤ 5) signaled in slice
header
Skip mode = merge mode + no residual
– Advanced Motion Vector Prediction (AMVP)
Use spatial/temporal PUs to predict current MV

140
HEVC…
• Transforms
– Core transforms: DCT based
4x4, 8x8, 16x16, and 32x32
Square transforms only
Support partial factorization
Near-orthogonal
Nested transforms
– Alternative 4x4 DST
4x4 intra blocks, luma only
– Transform skipping mode
By-pass the transform stage
Most effective on “screen content”
4x4 TBs only

141
HEVC…
• Scaling and Quantization
– HEVC uses a uniform reconstruction quantization (URQ)
scheme controlled by a quantization parameter (QP).
– The range of the QP values is defined from 0 to 51

142
HEVC…
• Entropy Coding
– One entropy coder, CABAC
Reuse H.264 CABAC core algorithm
More friendly to software and hardware
implementations
Easier to parallelize, reduced HW area, increased
throughput
– Context modeling
Reduced # of contexts
Increased use of by-pass bins
Reduced data dependency
– Coefficient coding
Adaptive coefficient scanning for intra 4x4 and 8x8
▫ Diagonal upright, horizontal, vertical
Processed in 4x4 blocks for all TU sizes
Sign data hiding:
▫ Sign of first non-zero coefficient conditionally hidden in
the parity of the sum of the non-zero coefficient
magnitudes
▫ Conditions: 2 or more non-zero coefficients, and
“distance” between first and last coefficient > 3

143
HEVC…
• Entropy Coding - CABAC
– Binarization: CABAC uses Binary Arithmetic Coding which means that only binary decisions (1 or
0) are encoded. A non-binary-valued symbol (e.g. a transform coefficient or motion vector) is
"binarized" or converted into a binary code prior to arithmetic coding. This process is similar to the
process of converting a data symbol into a variable length code but the binary code is further
encoded (by the arithmetic coder) prior to transmission.
– Stages are repeated for each bit (or "bin") of the binarized symbol.
– Context model selection: A "context model" is a probability model for one or more bins of the
binarized symbol. This model may be chosen from a selection of available models depending on
the statistics of recently coded data symbols. The context model stores the probability of each bin
being "1" or "0".
– Arithmetic encoding: An arithmetic coder encodes each bin according to the selected probability
model. Note that there are just two sub-ranges for each bin (corresponding to "0" and "1").
– Probability update: The selected context model is updated based on the actual coded value (e.g. if
the bin value was "1", the frequency count of "1"s is increased)

144
HEVC…
• Parallel Processing Tools
– Slices
– Tiles
– Wavefront parallel processing (WPP)
– Dependent Slices
• Slices
– Slices are a sequence of CTUs that are processed in the order
of a raster scan. Slices are self-contained and independent
– Each slice is encapsulated in a separate packet

145
HEVC…
• Tile
– Self-contained and independently decodable rectangular regions
– Tiles provide parallelism at a coarse level of granularity
Tiles more than the cores  Not efficient  Breaks dependencies

146
HEVC…
• WPP
– A slice is divided into rows of CTUs. Parallel processing of rows
– The decoding of each row can be begun as soon a few decisions have
been made in the preceding row for the adaptation of the entropy coder.
– Better compression than tiles. Parallel processing at a fine level of
granularity.
No WPP with tiles !!

147
HEVC…
• Dependent Slices
– Separate NAL units but dependent (Can only be decoded after part of
the previous slice)
– Dependent slices are mainly useful for ultra low delay applications
Remote Surgery
– Error resiliency gets worst
– Low delay
– Good Efficiency  Goes well with WPP

148
HEVC…
• Slice Vs Tile
– Tiles are kind of zero overhead slices
Slice header is sent at every slice but tile information once for a sequence
Slices have packet headers too
Each tile can contain a number of slices and vice versa
– Slices are for :
Controlling packet sizes
Error resiliency
– Tiles are for:
Controlling parallelism (multiple core architecture)
Defining ROI regions

149
HEVC…
• Tile Vs WPP
– WPP
Better compression than tiles
Parallel processing at a fine level of granularity
But …
Needs frequent communication between processing units
If high number of cores Can’t get full utilization
– Good for when
Relatively small number of nodes
Good inter core communication
No need to match to MTU size
Big enough shared cache

150
HEVC…
• In-Loop Filters
– Two processing steps, a deblocking filter (DBF) followed by an
sample adaptive offset (SAO) filter, are applied to the
reconstructed samples
The DBF is intended to reduce the blocking artifacts due to block-
based coding
The DBF is only applied to the samples located at block
boundaries
The SAO filter is applied adaptively to all samples satisfying
certain conditions. e.g. based on gradient.

151
HEVC…
• Loop Filters: Deblocking
– Applied to all samples adjacent to a PU or TU boundary
Except the case when the boundary is also a picture boundary, or
when deblocking is disabled across slice or tile boundaries
– HEVC only applies the deblocking filter to the edge that are
aligned on an 8×8 sample grid
This restriction reduces the worst-case computational complexity
without noticeable degradation of the visual quality
It also improves parallel-processing operation
– The processing order of the deblocking filter is defined as
horizontal filtering for vertical edges for the entire picture first,
followed by vertical filtering for horizontal edges.

152
HEVC…
• Loop Filters: Deblocking
– Simpler deblocking filter in HEVC (vs H.264 )
– Deblocking filter boundary strength is set according to
Block coding mode
Existence of non zero coefficients
Motion vector difference
Reference picture difference

153
HEVC…
• Loop Filters: SAO
– A process that modifies the decoded
samples by conditionally adding an
offset value to each sample after the
application of the deblocking filter,
based on values in look-up tables
transmitted by the encoder.
– SAO: Sample Adaptive Offsets
New loop filter in HEVC
Non-linear filter
– For each CTB, signal SAO type and
parameters
– Encoder decides SAO type and
estimates SAO parameters (rate-
distortion opt.)

154
HEVC…
• Special Coding
– I_PCM mode
The prediction, transform, quantization and entropy coding are bypassed
The samples are directly represented by a pre-defined number of bits
Main purpose is to avoid excessive consumption of bits when the signal
characteristics are extremely unusual and cannot be properly handled by hybrid
coding
– Lossless mode
The transform, quantization, and other processing that affects the decoded picture
are bypassed
The residual signal from inter- or intrapicture prediction is directly fed into the
entropy coder
It allows mathematically lossless reconstruction
SAO and deblocking filtering are not applied to this regions
– Transform skipping mode
Only the transform is bypassed
Improves compression for certain types of video content such as computer-
generated images or graphics mixed with camera-view content
Can be applied to TBs of 4×4 size only

155
HEVC…
• High Level Parallelism
– Independently decodable packets
– Sequence of CTUs in raster scan
– Error resilience
– Parallelization
– Independently decodable (re-entry)
– Rectangular region of CTUs
– Parallelization (esp. encoder)
– 1 slice = more tiles, or 1 tile = more slices
– Rows of CTUs
– Decoding of each row can be parallelized
– Shaded CTU can start when gray CTUs in
row above are finished
– Main profile does not allow tiles + WPP
combination

156
HEVC…
• Profiles, Levels and Tiers
– Historically, profile defines collection of coding
tools, whereas Level constrains decoder
processing load and memory requirements
– The first version of HEVC defined 3 profiles
Main Profile: 8-bit video in YUV4:2:0 format
Main 10 Profile: same as Main, up to 10-bit
Main Still Picture Profile: same as Main, one
picture only
– Levels and Tiers
Levels: max sample rate, max picture size,
max bit rate, DPB and CPB size, etc
Tiers: “main tier” and “high tier” within one
level

157
HEVC…
• Complexity Analysis
– Software-based HEVC decoder capabilities
(published by NTT Docomo)
Single-threaded: 1080p@30 on ARMv7
(1.3GHz),1080p@60 decoding on i5
(2.53GHz)
Multi-threaded: 4Kx2K@60 on i7 (2.7GHz),
12Mbps, decoding speed up to 100fps
– Other independent software-based HEVC
real-time decoder implementations published
by Samsung and Qualcomm during HEVC
development
– Decoder complexity not substantially higher
More complex modules: MC, Transform, Intra
Pred, SAO
Simpler modules: CABAC and deblocking

158
HEVC…
• Quality Performance

VIDEO CODECS

Recommandé

Recommandé

Contenu connexe

Tendances

Tendances (20)

Similaire à VIDEO CODECS

Similaire à VIDEO CODECS (20)

Plus de Vinayagam Mariappan

Plus de Vinayagam Mariappan (7)

Dernier

Dernier (20)

VIDEO CODECS