CaseStudy_ACS.pdf contains a transparent image overlaying the entire page.
This overlaying transparent image fools TreeExtractor into thinking it is scanned.
output = pdftotree.parse("tests/input/CaseStudy_ACS.pdf", favor_figures="True")
assert output.count("ocrx_word") == 1 // single appearance in ocr-capabilities
output = pdftotree.parse("tests/input/CaseStudy_ACS.pdf", favor_figures="False")
assert output.count("ocrx_word") >= 1000
After Change
This overlaying transparent image fools TreeExtractor into thinking it is scanned.
output = pdftotree.parse("tests/input/CaseStudy_ACS.pdf")
soup = BeautifulSoup(output)
assert len(soup.find_all(class_="ocrx_word")) >= 1000
assert len(soup.find_all("figure")) == 3
// Adapted from https://github.com/ocropus/hocr-tools/blob/v1.3.0/hocr-check
def get_prop(node: Tag, name: str) -> Optional[str]:
title = node.get("title")
if not title:
return None
props = title.split(";")
for prop in props:
(key, args) = prop.split(None, 1)
if key == name:
return args
return None
// Adapted from https://github.com/ocropus/hocr-tools/blob/v1.3.0/hocr-check
def get_bbox(node: Tag) -> box:
bbox = get_prop(node, "bbox")
if not bbox:
return None
return box(*[int(x) for x in bbox.split()])
// Check if words are extracted even though they are overlapped by a figure (/ኩ).
page = soup.find(class_="ocr_page") // checking only 1st page is good enough.
words = [get_bbox(word) for word in page.find_all(class_="ocrx_word")]
figure = get_bbox(page.find("figure"))
assert all([figure.contains(word) for word in words])
