pdf-figure-extraction

Extract figures from born-digital PDFs and render in JATS XML

Assumes we have pdf2xml available (a MacOS X executable is provided here) and a RIS file for a reference that includes ISSN, volume, space, and a URL for the PDF.

Attempts to extract text and images from PDF and extract figures and associated captions, creating a simple JATS XML file summarising article metadata and including figures. This is rendered into HTML using a XSL stylesheet.

Next step would be processing JATS XML into JSON suitable for uploading to Zenodo and BLR.

Journals

Journal specific things:

• alignment of figure caption and image
• whether we need to filter out text that overlaps with image

Works

• Acta Phytotaxonomica et Geobotanica 1346-7565
• Arnaldoa 1815-8242
• Austrobaileya 0155-4131
• Blumea 0006-5196
• Botanical Studies (Taipei) 1817-406
• Garden’s Bulletin Singapore 0374-7859
• Kew Bulletin 1874-933X paywall
• Lankesteriana 1409-3871
• Muelleria 0077-1813
• Philippine Journal of Science 0031-7683
• Phytologia 0031-9430
• Proceedings of The California Academy of Sciences 0068-547X
• Raffles Bulletin of Zoology 0217-2445
• South African Journal of Botany (Elsevier open access) 0254-6299
• Taiwania 0372-333X (issues using curl to get file, need user-agent)
• Teleopea 0312-9764

Fails

• Annales Botanici Fennici 0003-3847 (some captions are to the side of the image, need to be clever about this).
• Australian Journal of Entomology 1326-6756 complex multipart figures that overlap with both images and text. Interestingly Plazi has managed to process this, e.g. https://zenodo.org/record/269133
• Bothalia (complex multipart pictures) 0006-8241
• Bull. Bot. Res., Harbin (Chinese text, images not extracted by pdftoxml) 1673-5102
• Candollea 0373-2967 (figure captions small and left aligned, current code doesn’t find figs 3 and 4)
• Contributions To Natural History (1660-9972) all seem to be combined text plus images
• Integrative Biosciences (1738-6357) the text comes out with lots of spaces, and the Fig. title is a separate text block from caption(!). Need to handle text with spaces.
• Memoirs of Museum Victoria (1447-2554) line drawings don’t come out as images!? We would need to find large chunks of white space, or learn how PDF stores line drawings.
• Nordic Journal 0107-055X (early issue) PDF is OCR’d so need to extract images from page.
• Plants 2223-7747 (PDF seems complex, but we also have JATS XML)
• Records of the Australian Museum 0067-1975 DOI:10.3853/j.2201-4349.67.2015.1646 has two figures where the caption overlaps the figure.
• Smithsonian Contributions to Zoology 0081-0282 DOI:10.5479/si.00810282.636.1 got most but some figures are rotated with captions rotated as well :(
• Thai Forest Bulletin (Botany) 0495-3843 (Composite figure causes problems, e.g. S0495-38432016004400128 )