TableExtraction

This is a tool to extract tables in document images. The method is composed of 6 steps:

1. Line segment detection
2. Horizontal and vertical segment filtering
3. Line segment recovery
4. Suppression of segments belonging to text
5. Table cell extraction
6. Table reconstruction

Quick setup

• Requires OpenCv library.
• CMakeLists.txt provided for cmake.

To install the program see Install.txt file

Examples

Result (with intermediate steps) for eu-002_page0.png:

./TableExtraction -i ../Samples/eu-002_page0.png -o eu-002_page0_res.png

Input image	FBSD detector	Filtering and recovering segments
Removing text segments	Table extraction	Output image

Result for eu-001_page0.png:

./TableExtraction -i ../Samples/eu-001_page0.png -o eu-001_page0_res.png

Input image

Output image

Result for us-001_page0.png:

./TableExtraction -i ../Samples/us-001_page0.png -o us-001_page0_res.png

Input image

Output image

Result for 1_301.jpg:

./TableExtraction -i ../Samples/1_301.jpg -o 1_301_res.png

Input image

Output image

Limit cases

Result for 10.1.1.1.2111_7.jpg:

./TableExtraction -i ../Samples/10.1.1.1.2111_7.jpg -o 10.1.1.1.2111_7_res.png

Input image

Output image (graphics are mistakenly recognized as tables)

Result for 1078_082.png:

./TableExtraction -i ../Samples/1078_082.png -o 1078_082_res.png

Input image

Output image (Boundless table non detected)