Oct 21, 2008 IIP2008 1

Image Segmentation of Historical Handwriting from Palm Leaf

Manuscripts

Olarik SurintaMahasarakham University

Thailand

Oct 21, 2008 IIP2008 2

Introduction

• Palm leaf manuscripts have been a popular written media for over a thousand years in Southeast Asia

• Palm leaves were used for recording the history, knowledge and local wisdoms such as– Medical treatments– Buddhist doctrine– The story of dynasties

Oct 21, 2008 IIP2008 3

Introduction (cont)

• Mahasarakham University is establishing Palm Leaf Manuscript Preservation Project for the discovery, preservation and protection of palm leaf manuscripts from Northeast Thailand

palm leaf manuscript

Oct 21, 2008 IIP2008 4

Proposed framework

• We use palm leaf manuscripts consisting of 227 pages to do research work

• The system processes consist of– Background elimination– Line segmentation, and– Character segmentation

Oct 21, 2008 IIP2008 5

Proposed framework (cont)

Framework of the BILAN (palm leaf manuscripts) system

Oct 21, 2008 IIP2008 6

Convert Image from RGB color to Grey Image

• We use this equation to convert RGB color to Grey image

Y = 0.3R + 0.59G + 0.11BRGB color

Grey image

Oct 21, 2008 IIP2008 7

Noise Reduction

• Noise is maybe appearing from the scanning process.

• This process is removing noise from Grey image using Gaussian Filtering

grey image before and after noise reduction

Oct 21, 2008 IIP2008 8

Background Elimination using Otsu’s Algorithm

• This method proposed by Otsu. It based on grey level histogram

Otsu’s threshold value method

Oct 21, 2008 IIP2008 9

Background Elimination using Otsu’s Algorithm (cont)

Grey image

binary image after background elimination

Oct 21, 2008 IIP2008 10

Image recovery

• We apply Mathematical Morphology in this research such as– Dilation– erosion

binary image binary image after morphology

Oct 21, 2008 IIP2008 11

Line Segmentation

• Projection profile analysis is a popular technique for line segmentation.

• We use horizontal projection profile analysis because the texts in most document images are aligned along horizontal lines

Oct 21, 2008 IIP2008 12

Line Segmentation (cont)

Line segmentation histogram

Image after line segmentation

Oct 21, 2008 IIP2008 13

Line Segmentation (cont)

Oct 21, 2008 IIP2008 14

Character Segmentation

• In this step, use vertical projection profile analysis.

• we apply a threshold value on the length of the space in between the characters

image after vertical projection profile

Oct 21, 2008 IIP2008 15

Experimental Results

• The method was tested using a set of 227 palm leaf manuscripts

Oct 21, 2008 IIP2008 16

Background Elimination Results

Background Elimination Accuracy

Number of documents Percentage of background elimination segmented

Complete 138 61

Incomplete 89 39

Total 227 100

Oct 21, 2008 IIP2008 17

Line Segmentation Results

Number of Segmented Lines

Percentage of lines correctly segmented

4 78%

5 87%

Average 82.5%

Oct 21, 2008 IIP2008 18

Complete background elimination

Oct 21, 2008 IIP2008 19

Incomplete background elimination

Oct 21, 2008 IIP2008 20

Incomplete background elimination

Oct 21, 2008 IIP2008 21

Future work

• Application this research for OCR system.• Translation Palm leaf manuscripts into Thai

language.

Oct 21, 2008 IIP2008 22

End of this presentation

Thank you very much