Extraction and segmentation of tables from Chinese ink documents based on a matrix model

1
Extraction and segmentation of tables from
Chinese ink documents based on a matrix model

• Zhang Xi-Wen
• CSE, CUHK and HCI Lab., ISCAS
• 2005.10.24

2
Outline

• 1 Tables in an ink document.
• 2 A matrix for an ink document.
• 3 Ink tables are extracted and segmented.
• 4 Experimental results.
• 5 Conclusion.

3
Ink documents

• Ink documents are produced by digital ink
  capturers.
• Many objects are contained in an ink document.
• There are many components in an ink table.

4
1.1 Objects in an ink document

• Strokes.
• Objects.

5
Text

• Paragraph.
• Text-line, Expression.
• Character, Word, Symbols.

6
Graphics

• Long.
• Parts of tables and flowcharts.

7
Table

• Text (simple).
• Graphics.
• Bordering lines.
• Separating lines.

8
1.2 Components in an ink table

• Strokes.
• Row, Column.
• Header.
• Cell.
• Sub-header.
• Caption.
• Lines.

9
1.3 Our approach

• Previous approaches.
• A matrix model.

10
2 A matrix for a ink document

• Components in an ink document are extracted.
• An ink document can be modeled be a matrix.

11
2.1 Ink components

• An ink character.
• An ink line.
• An ink row.

12
2.2 Extract components in an ink document

• Ink characters.
• Ink lines.
• Ink rows.

13
(No Transcript)
14
2.2 A matrix model

• Multiple levels.
• Context.

15
3 Ink tables are extracted and segmented

• Extraction.
• Segmentation.

16
3.1 Table extraction

• An identical distribution of writing lines.
• The same drawing rows (if available) associated.

17

• A seed-table.
• The same distribution.
• The seed-table grows.

18
3.2 Table segmentation

• Rows.
• Columns.
• Headers.
• Cells.

19
An segmented ink table is modified and recognized.
20
(No Transcript)
21
4 Experimental results and performance analyses
22
4.1 Experimental results
23
(No Transcript)
24
4.2 performance analyses

• Strokes, captions, headers, cells, rows, and
  columns.
• The precision rate and the recall rate.

25
(No Transcript)
26
4.3 performance comparison

• Quality.
• Quantity.

27
Quality comparison
28
(No Transcript)
29
5 Conclusion

• A matrix model for extracting and segmenting ink
  tables.
• More ink tables can be processed.
• Extracted ink tables are decomposed.

30

• Thank you very much for
• your criticism, comments and suggestions!
• Email xwzhang_at_cse.cuhk.edu.hk
• Tel [phone]