DPCM - Overview

n Principle of Differential Pulse Code Modulation (DPCM)

n Characteristics of DPCM quantization errors
n Adaptive intra-interframe DPCM
n Conditional Replenishment

Bernd Girod: EE368b Image and Video Compression DPCM no. 1

Principle of DPCM
s + e e'
input entropy channel
quantizer coder
-
+
coder
+
s
predictor s'

s' + e'
entropy channel
output
decoder
+
decoder
predictor s

Prediction error Reconstruction Reconstruction error =

quantization error

e = s − sˆ s ′ = e ′ + sˆ s ′ − s = e′ − e = q

Bernd Girod: EE368b Image and Video Compression DPCM no. 2

1
 Quantization error feedback
in the DPCM coder
n Assuming a linear predictor, the DPCM coder is equivalent
to the following structure:
~
s + e e e'
quantizer
- -
-
s (s) s (q)
predictor predictor
+
q (e)
n Transfer function of the prefilter: Ω - abbreviation for
frequency vector,
e.g., (ω x ,ω y )
E˜ (Ω) = [1 − P(Ω)]S(Ω)

transfer function of the predictor

n Transfer function of quantization error feedback:

E ′ (Ω) = E˜ (Ω) + [1 − P(Ω)]Q(Ω)

Bernd Girod: EE368b Image and Video Compression DPCM no. 3

Power spectrum of the DPCM

quantization error
n Power spectral density of the quantization error q measured
for intraframe DPCM with a 16 level quantizer

ωx
−π 0
ωy 0
π −π

Bernd Girod: EE368b Image and Video Compression DPCM no. 4

2
 Signal distortions due to
intraframe DPCM coding
n Granular noise: random
noise in flat areas of the
picture
n Edge busyness: jittery
appearance of edges
(for video)
n Slope overload: blur of
high-contrast edges,
Moire patterns in periodic
structures.

Bernd Girod: EE368b Image and Video Compression DPCM no. 5

Example of intraframe DPCM coding

1 bit/pixel 2 bit/pixel 3 bit/pixel

prediction error coding edge busyness

slope overload granular noise n Linear predictor:

0 1/4 1/4

1/2

n Lloyd-Max quantizers
4 bit/pixel original n Fixed-length coding

Bernd Girod: EE368b Image and Video Compression DPCM no. 6

3
 Interframe coding of video signals

n Interframe coding exploits:

l similarity of temporally successive pictures
l temporal properties of human vision
n Important interframe coding methods:
l Adaptive intra-interframe coding
l Conditional replenishment
l Motion-compensated prediction
l Motion-compensated interpolation

Bernd Girod: EE368b Image and Video Compression DPCM no. 7

Principle of adaptive intra-interframe DPCM

n Predictor is switched between two states:

A: Intraframe prediction for B: Interframe prediction (previous frame
moving or changed areas. prediction) for still areas of the picture.

S 22 S23 S24
framems
40
interval τ
S22 S 23 S24
S2 S3 S4 S2 S3 S4
S 21 S20 S25

FRAME N - 1
S21 S 20 S25
S1 S0 S1 S0
FRAME N - 1
FRAME N FRAME N

Sˆintra = a1 S1 + a 2 S2 + a3 S3 + a 4 S4 Sˆinter = S20

Bernd Girod: EE368b Image and Video Compression DPCM no. 8

4
 Intra-interframe DPCM: feedback adaptation

s + e e' Variable Variable

e' s'
Quantizer
length coder length decoder
-
^ ^s
s
^ inter
s
^ inter
s Interframe
Interframe predictor
predictor s'
Intraframe
Intraframe predictor
predictor ^intra
s
^ intra
s
Predictor
Predictor adaptation
adaptation

Coder Decoder

Bernd Girod: EE368b Image and Video Compression DPCM no. 9

Intra-interframe DPCM: feedforward adaptation

s + e e' e' s'

Quantizer Variable Variable
length coder length decoder
-
^s ^s

^
s
^s inter Interframe
inter
Interframe predictor
predictor s'
Intraframe
Intraframe ^ predictor
predictor s
s^intra intra

Predictor
adaptation

intra-/interframe
switching information

Coder Decoder

Bernd Girod: EE368b Image and Video Compression DPCM no. 10

5
 Conditional replenishment
TRANSMISSION SIGNAL
CODING, CHANNEL BUFFERING, OUTPUT
ADDRESSING, DECODING,
SIGNAL
INPUT
BUFFERING ADDRESSING

SEGMENTER
Change
(MOVEMENT
DETECTOR)
detector

FRAME DELAY FRAME DELAY

Frame
(1 PICTUREstore
MEMORY) (1Frame
PICTUREstore
MEMORY)

CODER DECODER

n Still areas: repeat from frame store

n Moving areas: encode and transmit address and waveform

Bernd Girod: EE368b Image and Video Compression DPCM no. 11

Change detection
n Example of a pixel-wise change detector
Current
frame

+ Average Eliminate isolated

ABS of 3x3 points or pairs of
window points
Threshold Decision
- changed/
unchanged
Previous
frame

n Example of a block-wise change detector

Current
frame
+
Accumulate
ABS over NxN
blocks Decision
- Threshold
changed/
unchanged
Previous
frame

Bernd Girod: EE368b Image and Video Compression DPCM no. 12

6
 The “Dirty Window” effect

n Conditional replenishment scheme with change detection

threshold set too high leads to the subjective impression of
looking through a dirty window.

Background
Moving area
picked up by Moving areas
change detector missed by
change detector

Bernd Girod: EE368b Image and Video Compression DPCM no. 13

Crawford noise reduction filter

δ)
f(δ

+ δ Non- δ)
f(δ δ
NL
linearity
noisy video -
signal

+ +

frame
store clean video
signal

Bernd Girod: EE368b Image and Video Compression DPCM no. 14

7
 DPCM - Summary

n DPCM: Prediction from previously coded/transmitted samples (known at

transmitter and receiver)
n Typical signal distortions for intraframe DPCM: granular noise, edge
busyness, slope overload
n Adaptive Intra-Interframe-DPCM: forward adaptation vs. backward
adaptation
n Conditional replenishment: only transmit frame-to-frame changes
n Temporal noise reduction by nonlinear, recursive frame differencing

Bernd Girod: EE368b Image and Video Compression DPCM no. 15