TW015 Ting Yu Chang Advisor: Chi Cheng Jou Institute of - - PowerPoint PPT Presentation

tw015 ting yu chang advisor chi cheng jou
SMART_READER_LITE
LIVE PREVIEW

TW015 Ting Yu Chang Advisor: Chi Cheng Jou Institute of - - PowerPoint PPT Presentation

TW015 Ting Yu Chang Advisor: Chi Cheng Jou Institute of Electrical Control Engineering NCTU July 11 th ,2015 What What is word segmentation? Why Why do we need word segmentation? How How do we implement word


slide-1
SLIDE 1

‐TW015‐ Ting‐Yu Chang Advisor: Chi‐Cheng Jou

Institute of Electrical Control Engineering NCTU

July 11th ,2015

slide-2
SLIDE 2
  • What

What is word segmentation?

  • Why

Why do we need word segmentation?

  • How

How do we implement word segmentation?

  • Conclusions

Conclusions

slide-3
SLIDE 3

What What is word segmentation?

slide-4
SLIDE 4
slide-5
SLIDE 5

Segmentation is the process of partitioning a digital image into multiple segments

slide-6
SLIDE 6
slide-7
SLIDE 7

Handwriting here

scan

slide-8
SLIDE 8

Why Why

do we need word segmentation?

slide-9
SLIDE 9

I can do it!

slide-10
SLIDE 10
slide-11
SLIDE 11

How How do we implement word segmentation?

slide-12
SLIDE 12

For the software:

 NIOS II  HPS(Hard Processor System)

Run Histogram equalization

 NIOS II: 2.5 seconds  HPS:

0.065 seconds

HPS is Faster!!

For the hardware:

slide-13
SLIDE 13
slide-14
SLIDE 14
slide-15
SLIDE 15
slide-16
SLIDE 16
slide-17
SLIDE 17

Normalize Normalize Normalize Normalize

slide-18
SLIDE 18

Calculating minimum/maximum distance between every neighbors

score=0.7minimum distance + 0.3maxmum distance

Merge parts to the nearest neighbor

slide-19
SLIDE 19

Reduced to 2 min per picture(3 times improvement) It is too SLOW!!(6 mins per picture)

Sobel edge detection

Reduced to 10 seconds !!!

Reduce to only 1 point for each partition

6mins → 2mins→ 10 seconds

slide-20
SLIDE 20

Conclusions Conclusions

slide-21
SLIDE 21

What What is word segmentation?

 For Chinese words  For hand‐writing

Why Why do we need word segmentation?

 For massive documents  Digital modification: sorting, painting and deleting unwanted word

How How do we implement word segmentation?

 Use FPGA and ARM  Hardware issues: capture image data from vga output  Software issues: merge incorrectly segmented parts  Performance: 10 seconds per image  Accuracy: 85%↑

slide-22
SLIDE 22

Thanks for your attention Thanks for your attention