JorjMcKie
commented
3 months ago The reason is that the PDF only has a handful of pages containing images at all: (0-based): 11, 149, 154, 158, 162, 164.
Everything else are vector graphics. They cannot be extracted. All you can do is to find suitable rectangles wrapping them, then make a pixmap of those rectangles and save them as an image file.
We have detection algorithms that are able to identify / cluster atomic graphical commands to larger drawings - see Page method cluster_drawings. The method returns a list of rectangles containing such cluster.
Use this list as described above. There also is a little utility doing this here.
As per header / footer exclusion:
This is your job - PDF knows nothing about what these things are. You could do the obvious thing and ignore every bbox returned by page.cluster_drawings() which has y0 <= 72 or y1 >= page.rect.height - 72.
Modify 72 as appropriate.
Addresses #3539
We previously did not detect dotted lines used as table gridlines.
If one of width / height is LE edge_min_length and LT the other dimension, the rectangle is treated as a vertical / horizontal line. We incorrectly used the dimension-specific snap values for this.
We also no longer ignore rectangles if both, width and height are smaller than edge_min_length, but leave this to the snapping and joining algorithms further down the road.