def get_text(self):
"""Does OCR on this image."""
image_writer = ImageWriter("temp")
try:
temp_image = image_writer.export_image(self._image_obj)
except PDFNotImplementedError:
# No filter method available for this stream
# https://github.com/euske/pdfminer/issues/99
return u""
try:
text = image_to_string(Image.open("temp/" + temp_image),
lang="fin")
except IOError:
# PdfMiner did not return an image
# Let's try to create one ourselves
# TODO: Create proper color_mode values from ColorSpace
# Most of the times "L" will create something good enough
# for OCR, though
temp_image = Image.frombuffer("L",
self._image_obj.srcsize,
self._stream.get_data(), "raw",
"L", 0, 1)
text = image_to_string(temp_image, lang="fin")
unlink("temp/" + temp_image)
return text
def parse_img(img_obj, labels):
from pdfminer.image import ImageWriter
from subprocess import call
from pdfminer.pdftypes import PDFNotImplementedError
from os import remove, rename, path
try:
logging.info("Writing " + img_obj.name)
image_writer = ImageWriter(".")
# TODO this is not thread safe... (so run with "-p 1" for now)
output_filename = image_writer.export_image(img_obj)
# TODO rename(output_filename, ...)
logging.info("Written " + output_filename)
logging.info("Calling Tesseract hOCR")
call(["tesseract", output_filename, "out", "-l", "nld", "hocr"])
# TODO remove(output_filename)
if path.isfile("./out.html"):
# TODO fix race condition
from extracthocr import get_bbox_page, get_bbox_texts
for label in get_bbox_texts("./out.html"):
labels.append(convert_label_bbox(label, img_obj.bbox, get_bbox_page("./out.html")))
remove("./out.html")
else:
logging.error("Image object kon niet verwerkt worden door Tesseract: " + output_filename + ".")
except PDFNotImplementedError as e:
logging.error(
"Image object kon niet verwerkt worden: "
+ str(e)
+ ". Mogelijk helpt het opslaan vam de PDF met ondersteuning voor Acrobat 4.x en later."
)
def find_object_recursively(layout_obj):
# LTTextBoxを継承するオブジェクトの場合は1要素のリストを返す。
# print("obj_name:", layout_obj.__class__.__name__)
if isinstance(layout_obj, LTTextBox):
return [layout_obj]
if isinstance(layout_obj, LTImage):
# print("LTImage:", layout_obj.__class__.__name__)
imagewriter = ImageWriter(output_image_dir)
imagewriter.export_image(layout_obj)
# TODO jpeg以外の画像を抽出したい（PDFMinerでは不可ぽい）
# stream = child_fig.stream
# filters = stream.get_filters()
# (width, height) = child_fig.srcsize
# print("width, height, len(filters), filters[0][0]:", width, height, len(filters), filters[0][0])
# ext = '_.%d.%dx%d.img' % (child_fig.bits, width, height)
# name = child_fig.name + ext
# print(child_fig.bits, name, child_fig.colorspace)
# if filters[0][0] in LITERALS_FLATE_DECODE:
# with open('outputImg/' + name, "wb") as fp:
# # imgdata = zlib.decompress(stream.data)
# data = stream.get_data()
# fp.write(data)
# fp.close()
return [layout_obj]
if isinstance(layout_obj, LTLine):
return [layout_obj]
if isinstance(layout_obj, LTRect):
return [layout_obj]
# LTLine も LTRect も、LTCurve を継承しているので、これで両方とも返る
# if isinstance(layout_obj, LTCurve):
# return [layout_obj]
# LTPageやLTFigureなど、LTContainerを継承するオブジェクトは子要素を含むので、再帰的に探す。
# if isinstance(layout_obj, LTPage):
if isinstance(layout_obj, LTContainer):
boxes = []
for child_con in layout_obj:
boxes.extend(find_object_recursively(child_con))
return boxes
# print("none object:", layout_obj.__class__.__name__)
return [] # その他の場合は空リストを返す。
def get_text(self):
"""Does OCR on this image."""
image_writer = ImageWriter("temp")
try:
temp_image = image_writer.export_image(self._image_obj)
except PDFNotImplementedError:
# No filter method available for this stream
# https://github.com/euske/pdfminer/issues/99
return u""
try:
text = image_to_string(Image.open("temp/" + temp_image),
lang="swe")
except IOError:
# PdfMiner did not return an image
# Let's try to create one ourselves
# TODO: Create proper color_mode values from ColorSpace
# Most of the times "L" will create something good enough
# for OCR, though
temp_image = Image.frombuffer("L", self._image_obj.srcsize,
self._stream.get_data(), "raw", "L",
0, 1)
text = image_to_string(temp_image, lang="swe")
unlink("temp/" + temp_image)
return text
def get_html_tree(self) -> str:
# Create a temp folder where images are temporarily saved.
dirname = tempfile.mkdtemp()
imagewriter = ImageWriter(dirname)
doc = Document()
self.doc = doc
html = doc.createElement("html")
doc.appendChild(html)
head = doc.createElement("head")
html.appendChild(head)
# meta
meta = doc.createElement("meta")
head.appendChild(meta)
meta.setAttribute("name", "ocr-system")
meta.setAttribute("content",
f"Converted from PDF by pdftotree {__version__}")
meta = doc.createElement("meta")
head.appendChild(meta)
meta.setAttribute("name", "ocr-capabilities")
meta.setAttribute("content",
"ocr_page ocr_table ocrx_block ocrx_line ocrx_word")
meta = doc.createElement("meta")
head.appendChild(meta)
meta.setAttribute("name", "ocr-number-of-pages")
meta.setAttribute("content", f"{len(self.elems.keys())}")
# body
body = doc.createElement("body")
html.appendChild(body)
for page_num in self.elems.keys(): # 1-based
boxes: List[Tuple[str, float, float, float, float]] = []
for clust in self.tree[page_num]:
for (pnum, pwidth, pheight, top, left, bottom,
right) in self.tree[page_num][clust]:
boxes += [(clust.lower().replace(" ", "_"), top, left,
bottom, right)]
page = doc.createElement("div")
page.setAttribute("class", "ocr_page")
page.setAttribute("id", f"page_{page_num}")
width = int(self.elems[page_num].layout.width)
height = int(self.elems[page_num].layout.height)
page.setAttribute(
"title",
f"bbox 0 0 {width} {height}; ppageno {page_num-1}",
)
body.appendChild(page)
# TODO: We need to detect columns and sort acccordingly.
boxes.sort(key=cmp_to_key(column_order))
for box in boxes:
if box[0] == "table":
table = box[1:] # bbox
table_element = self.get_html_table(table, page_num)
page.appendChild(table_element)
elif box[0] == "figure":
elems: List[LTTextLine] = get_mentions_within_bbox(
box, self.elems[page_num].figures)
fig_element = doc.createElement("figure")
page.appendChild(fig_element)
top, left, bottom, right = [int(i) for i in box[1:]]
fig_element.setAttribute(
"title", f"bbox {left} {top} {right} {bottom}")
for img in [img for elem in elems for img in elem]:
if not isinstance(img, LTImage):
continue
filename = imagewriter.export_image(img)
with open(os.path.join(dirname, filename), "rb") as f:
base64 = b64encode(f.read()).decode("ascii")
if filename.endswith("jpg"):
mediatype = "jpeg"
elif filename.endswith("bmp"):
mediatype = "bmp"
else:
logger.info(
f"Skipping an unknown type image: {filename}.")
continue
logger.info(
f"Embedding a known type image: {filename}.")
img_element = doc.createElement("img")
fig_element.appendChild(img_element)
img_element.setAttribute("title", bbox2str(img.bbox))
img_element.setAttribute(
"src", f"data:image/{mediatype};base64,{base64}")
else:
element = self.get_html_others(box[0], box[1:], page_num)
page.appendChild(element)
return doc.toprettyxml()