VLSI Signal Processing Lab, Institute of Electronics National Chiao Tung University, Hsinchu, Taiwan...

27
2007/07/19 V L S I S i g n a l P r o c e s s i n g L a b , I n s t i t u t e o f E l e c t r o n i c s N a t i o n a l C h i a o T u n g U n i v e r s i t y , H s i n c h u , T a i w a n Platform-Based Design Group Segmentation with raster sliding local window Nelson Chang Institute of Electronics, National Chiao Tung University
  • date post

    20-Dec-2015
  • Category

    Documents

  • view

    223
  • download

    3

Transcript of VLSI Signal Processing Lab, Institute of Electronics National Chiao Tung University, Hsinchu, Taiwan...

2007/07/19

VL

SI S

ign

al P

roc

es

sin

g L

ab

, Ins

titute

of E

lec

tron

ics

Na

tion

al C

hia

o T

un

g U

niv

ers

ity, H

sin

ch

u, T

aiw

an

Platform-Based Design Group

Segmentation with raster sliding local window

Nelson ChangInstitute of Electronics,

National Chiao Tung University

2007/07/19 2

VL

SI S

ign

al P

roc

es

sin

g L

ab

Na

tion

al C

hia

o T

un

g U

niv

ers

ity, H

sin

ch

u, T

aiw

an

Platform-Based Design Group

Outline

• Stereo Matching with Segment• Traditional Watershed-based Segmentation• Downfall of tradition segmentation methods• Proposed Raster Sliding Local Window Method• Result• Future work

2007/07/19 3

VL

SI S

ign

al P

roc

es

sin

g L

ab

Na

tion

al C

hia

o T

un

g U

niv

ers

ity, H

sin

ch

u, T

aiw

an

Platform-Based Design Group

Census Diffusion WITHOUT Segmentation

2007/07/19 4

VL

SI S

ign

al P

roc

es

sin

g L

ab

Na

tion

al C

hia

o T

un

g U

niv

ers

ity, H

sin

ch

u, T

aiw

an

Platform-Based Design Group

Census Diffusion WITH segmentation

2007/07/19 5

VL

SI S

ign

al P

roc

es

sin

g L

ab

Na

tion

al C

hia

o T

un

g U

niv

ers

ity, H

sin

ch

u, T

aiw

an

Platform-Based Design Group

Mean Shift vs. Watershed

Mean Shift

Watershed

2007/07/19 6

VL

SI S

ign

al P

roc

es

sin

g L

ab

Na

tion

al C

hia

o T

un

g U

niv

ers

ity, H

sin

ch

u, T

aiw

an

Platform-Based Design Group

Traditional Watershed-based Segmentation

• Steps– Generate Gradient Map– Reconstruction/Global Denoise– Watershed flooding or Rain dropping– Region Merging

2007/07/19 7

VL

SI S

ign

al P

roc

es

sin

g L

ab

Na

tion

al C

hia

o T

un

g U

niv

ers

ity, H

sin

ch

u, T

aiw

an

Platform-Based Design Group

What makes an algorithm efficient to implement?

• Regular– Simple control– High data reusability

• Parallel– More scalable

• Small storage– Lower cost– More fast memory usage

state 0

state 1state 2

state 0

state 1state 2

state 3

state 4

PE PE PE PE PE

PE

PE

PE

4x speed up!!

Data reuse!!!

Simple FSM

Fast SRAM

Slow SDRAM

Temporary Data

Temporary Data

2007/07/19 8

VL

SI S

ign

al P

roc

es

sin

g L

ab

Na

tion

al C

hia

o T

un

g U

niv

ers

ity, H

sin

ch

u, T

aiw

an

Platform-Based Design Group

Traditional Watershed and Topoggan

• Watershed Flooding– Bottom-up

• Finds the minimum first

Irregular jumpy access pattern

HW: hard to reuseSW: low cache hit

2007/07/19 9

VL

SI S

ign

al P

roc

es

sin

g L

ab

Na

tion

al C

hia

o T

un

g U

niv

ers

ity, H

sin

ch

u, T

aiw

an

Platform-Based Design Group

Traditional Watershed and Topoggan

• Topoggan Rain Dropping– Top-down

• Finds the maximum peak first• Drop rain• Back trace the path

Back trace water flow path(tree traversal)

Irregular access start pointHW: less chance to reuse

SW: more cache missMore regular local access

pattern than watershed

High data dependencyHW: less parallelism

availableSW: not much impact

2007/07/19 10

VL

SI S

ign

al P

roc

es

sin

g L

ab

Na

tion

al C

hia

o T

un

g U

niv

ers

ity, H

sin

ch

u, T

aiw

an

Platform-Based Design Group

Our Goal

• Regular process and access pattern• Avoid tree traversal/tree traversal• Minimize iteration count

2007/07/19 11

VL

SI S

ign

al P

roc

es

sin

g L

ab

Na

tion

al C

hia

o T

un

g U

niv

ers

ity, H

sin

ch

u, T

aiw

an

Platform-Based Design Group

Overview (1/3)

• Flowchart

Read image information

gradient

Local-approach segmentation

start

Done

2007/07/19 12

VL

SI S

ign

al P

roc

es

sin

g L

ab

Na

tion

al C

hia

o T

un

g U

niv

ers

ity, H

sin

ch

u, T

aiw

an

Platform-Based Design Group

Gradient Generation (0th Iteration)

• Color space– YUV– Select the maximal gradient among the three components

• Gradient - Mathematical morphology – ABS(Dilation – Erosion)

• Dilation : Max in 3x3 square window

• Erosion : min in 3x3 square window

16 9 8

7 10 10

6 10 5

Gradient = 16-5 = 11

2007/07/19 13

VL

SI S

ign

al P

roc

es

sin

g L

ab

Na

tion

al C

hia

o T

un

g U

niv

ers

ity, H

sin

ch

u, T

aiw

an

Platform-Based Design Group

Local approachLocal

approach

Local de-noise

Basin/plateau?

Assign 3x3 window to

same regionDo nothing

Done image?

yes no

no

a

a

x has label?

N(x) ≥ x & N(x) has no label

Diffuse label upward

Do nothing

Done image?

Done

yes

yes no

yes

no

2007/07/19 14

VL

SI S

ign

al P

roc

es

sin

g L

ab

Na

tion

al C

hia

o T

un

g U

niv

ers

ity, H

sin

ch

u, T

aiw

an

Platform-Based Design Group

Local Denoise (1st Iteration)

i j k

l x m

n o p

Neighbor of x N(x) = {i, j, k, l, m, n, o, p}Gradient of x G(x)Gradient after local denoise G’

For each N(x), If | G(x) - G(N(x)) | < TH, G’(N(x) = G(x) Else G’(N(x)) = G(x)

19 24 82

37 30 38

10 13 12

G

19 30 82

30 30 30

10 13 12

G’

TH=10

2007/07/19 15

VL

SI S

ign

al P

roc

es

sin

g L

ab

Na

tion

al C

hia

o T

un

g U

niv

ers

ity, H

sin

ch

u, T

aiw

an

Platform-Based Design Group

An Example of Local Denoise

• G’ (Gradient after local denoise) of an overlapping pixel in two neighboring windows may be different

19 24 82

37 30 38

10 13 12

G

19 30 82

30 30 30

10 13 12

G’

TH=1042

10

29

24 82 38

38 38 10

10 13 38

G’

2007/07/19 16

VL

SI S

ign

al P

roc

es

sin

g L

ab

Na

tion

al C

hia

o T

un

g U

niv

ers

ity, H

sin

ch

u, T

aiw

an

Platform-Based Design Group

Adaptive Bi-threshold

• TH for the next window changes– Current window is a plateau/basin

• Next TH = TH_max

– Else• Next TH = TH_min

Plateau/Basin

Use TH_max

Possible Edge/Hill

Use TH_min

Current Next

2007/07/19 17

VL

SI S

ign

al P

roc

es

sin

g L

ab

Na

tion

al C

hia

o T

un

g U

niv

ers

ity, H

sin

ch

u, T

aiw

an

Platform-Based Design Group

Basin/Plateau Detection (1st Iteration)

Neighbor of x N(x) = {i, j, k, l, m, n, o, p}Gradient of x G(x)Gradient after local denoise G’Segment (Region) Label of x L(x)

For all N(x), If G’(N(x)) – G’(x) ≥ 0, L(x) = all L(N(x)) Else L(x) = no label

20 45 20

62 20 31

32 20 20

G’ L

No original label on N(x)

L

One N(x) already has labelBasin

2007/07/19 18

VL

SI S

ign

al P

roc

es

sin

g L

ab

Na

tion

al C

hia

o T

un

g U

niv

ers

ity, H

sin

ch

u, T

aiw

an

Platform-Based Design Group

Segment Label Aliasing (1st Iteration)

• For a basin/plateau

L

More than one N(x) already having different label

Set label alias

=

=

===

=

===

Label Alias Table

update

2007/07/19 19

VL

SI S

ign

al P

roc

es

sin

g L

ab

Na

tion

al C

hia

o T

un

g U

niv

ers

ity, H

sin

ch

u, T

aiw

an

Platform-Based Design Group

Label Upward Diffusion (2nd~Nth Iteration)

Neighbor of x N(x) = {i, j, k, l, m, n, o, p}Gradient of x G(x)Segment (Region) Label of x L(x)

For each N(x), If L(N(x)) = no label AND G(N(x)) – G(x) ≥ 0, L(N(x))=L(x) Else L(N(x)) = no label

19 24 82

37 30 38

10 13 12

G L

This can be done for more than 1 iterations

2007/07/19 20

VL

SI S

ign

al P

roc

es

sin

g L

ab

Na

tion

al C

hia

o T

un

g U

niv

ers

ity, H

sin

ch

u, T

aiw

an

Platform-Based Design Group

Label Replacement (N+1 Iteration)

• Update each pixel’s label– Use Label Alias Table– Similar to region (segment) merging

=

===

=

===

Label Alias Table

2007/07/19 21

VL

SI S

ign

al P

roc

es

sin

g L

ab

Na

tion

al C

hia

o T

un

g U

niv

ers

ity, H

sin

ch

u, T

aiw

an

Platform-Based Design Group

Segmentation with Raster Sliding Local Window

Watershed

Raster Sliding Window

2007/07/19 22

VL

SI S

ign

al P

roc

es

sin

g L

ab

Na

tion

al C

hia

o T

un

g U

niv

ers

ity, H

sin

ch

u, T

aiw

an

Platform-Based Design Group

Results

2007/07/19 23

VL

SI S

ign

al P

roc

es

sin

g L

ab

Na

tion

al C

hia

o T

un

g U

niv

ers

ity, H

sin

ch

u, T

aiw

an

Platform-Based Design Group

2007/07/19 24

VL

SI S

ign

al P

roc

es

sin

g L

ab

Na

tion

al C

hia

o T

un

g U

niv

ers

ity, H

sin

ch

u, T

aiw

an

Platform-Based Design Group

2007/07/19 25

VL

SI S

ign

al P

roc

es

sin

g L

ab

Na

tion

al C

hia

o T

un

g U

niv

ers

ity, H

sin

ch

u, T

aiw

an

Platform-Based Design Group

2007/07/19 26

VL

SI S

ign

al P

roc

es

sin

g L

ab

Na

tion

al C

hia

o T

un

g U

niv

ers

ity, H

sin

ch

u, T

aiw

an

Platform-Based Design Group

Ranking

• CDS stereo matching performance comparison on Middleburry website

2007/07/19 27

VL

SI S

ign

al P

roc

es

sin

g L

ab

Na

tion

al C

hia

o T

un

g U

niv

ers

ity, H

sin

ch

u, T

aiw

an

Platform-Based Design Group

Future work

• Complexity analysis – Algorithmic Complexity– Operation Complexity– SW execution– Storage Requirement

• Hardware implementation– ISCAS’09

• Refine performance– Boundary improvement– Label alias issue