Hi,
I'm trying to read PDFs and it looks like each character of each word is represented as individual features. Each PDF has thousands of these text features that I need to combine to make words. I'm stumped as to how to string these single character text features together to create words.
I have attached an example below.
Any help would be appreciated. Thanks in advance.
In the reader parameters, make sure the spatial text is read in as one feature per (text) block:
In theory, that should fix it, but my experience has been mixed, in particular with narrow fonts or particular page sizes. For some documents, the only reliable results I got was from this Python module, which is really exceptionally good at locating text blocks: https://pdfminersix.readthedocs.io/en/latest/ For maximum integration into your workspace, you'll have to implement it in a PythonCaller, but it's apparently also possible to use it on the command line.
Unfortunately, changing the Spatial Text parameter didn't help in my case and I haven't gotten around to trying to figure out the Python. Thanks tho for the suggestion. I'll give it try again later on.
Cheers
Unfortunately, changing the Spatial Text parameter didn't help in my case and I haven't gotten around to trying to figure out the Python. Thanks tho for the suggestion. I'll give it try again later on.
Cheers
As a starting point, here's sample code for the PythonCaller that will output a feature for each individual text block that is found within a PDF, as specified in the attribute "pdf_filename":
import fmeobjects
from pdfminer.high_level import extract_pages
from pdfminer.layout import LTTextBoxHorizontal, LAParams
class ExtractPdfTextBlocks(object):
def __init__(self):
# LAParams documentation can be found here:
# https://pdfminersix.readthedocs.io/en/latest/reference/composable.html
self.params = LAParams(line_margin=0,
detect_vertical=False,
boxes_flow=None,
all_texts=False)
def input(self, feature):
pdf_filename = feature.getAttribute('pdf_filename')
for page_layout in extract_pages(pdf_filename, laparams=self.params):
for element in page_layout:
if isinstance(element, LTTextBoxHorizontal):
text_string = str(element.get_text()).strip()
text_x_pos = element.x0
text_y_pos = element.y0
text_bbox = element.bbox
# Create a new feature for each text block found
new_feature = feature.clone()
new_feature.setAttribute('page_number', page_layout.pageid)
new_feature.setAttribute('text_x_pos', text_x_pos)
new_feature.setAttribute('text_y_pos', text_y_pos)
new_feature.setAttribute('text_bbox', str(text_bbox))
new_feature.setAttribute('text_string', text_string)
self.pyoutput(new_feature)
def close(self):
passYou'll want to expose the following attributes in the PythonCaller:
Sample output:
To install pdfminer.six, open a command prompt as local admin in the FME installation folder:
fme python -m pip install pdfminer.sixDocumentation for pdfminer.six: https://pdfminersix.readthedocs.io/en/latest/index.html