Distinguish table and vector graphic about pdf2docx

[dothinking]

One idea is to detect vector graphics, then exclude such paths from table parsing, but convert vector graphic (with calculated bbox) to pix-map with page.getPixmap(clip=bbox).

• extract all paths from pdf
• group/cluster paths into separate potential vector graphics regions
• check each group -> suppose a vector graphic if contains a certain number of curve/Bezier paths
  • if vector graphic: exclude these paths, but include bitmap in the same bbox
  • otherwise, include paths in this group for table parsing

Then two critical points:

1. extract paths according to PDF references
2. cluster paths and detect vector graphic

[dothinking]

cluster paths and detect vector graphic

It's equal to a disconnected GRAPH searching problem, and the critical point is building the adjacent list. In current implementation, brute force method is used O(n^2). It's acceptable (~2s) when the count of path is smaller than 2000.

In this case, building the adjacent list is a rectangle intersection problem -> to apply optimized algorithm next step.

[dothinking]

The following rectangle intersection solving algorithm was implemented.

A Rectangle-Intersection Algorithm with Limited Resource Requirements