[Python] Extracting decimal and negative numbers from PDF, where decimals look like exponents...

I am processing a set of PDF files using Python.
From each PDF I need to extract several monetary amounts and then write them to an Excel file.

The problem is that in the PDF the decimals are written with superscript digits after a comma, for example:

• 
  310,⁷⁶⁹⁷
  
  
• 
  −25,³⁴
  

These are normal decimal amounts (money), not powers.
I want to convert them to:

• 
  310.7697
  
  
• 
  -25.34
  

I am using:

• 
  PyMuPDF (fitz) to read the PDF text (and fall back to Tesseract OCR if needed),
  
  
• 
  some regex + string cleaning to extract the amounts,
  
  
• 
  openpyxl to write the results to Excel.
  

Below is a minimal reproducible example of the part that parses the amounts.
The text I show here (sample text) is similar to what I get after OCR:

import re

def clean_and_convert_amount(amount_str: str) -> float:
"""
Cleans a numeric string like:
- '1 068,²⁷⁵⁶'
- '1 068,2756'
- '−2,2072'
and converts it to a float, preserving:
- the minus sign
- decimal digits, even if they are written as superscript.
"""
if not amount_str:
return 0.0

raw = str(amount_str).strip()

# normalize different minus characters to '-'
raw = raw.replace('−', '-').replace('–', '-')

# if there is any '-' we consider the number negative
is_negative = '-' in raw

# map superscript digits -> normal digits
superscript_map = {
'⁰': '0', 'ⁱ': '1', '¹': '1', '²': '2', '³': '3',
'⁴': '4', '⁵': '5', '⁶': '6', '⁷': '7', '⁸': '8', '⁹': '9',
'\u00B2': '2', # ²
'\u00B3': '3', # ³
'\u00B9': '1', # ¹
}

def normalize_digit(ch: str) -> str:
if ch in superscript_map:
return superscript_map[ch]
if ch.isdigit():
return ch
return ''

# find decimal separator: comma (Romanian) or dot
sep_index = raw.find(',')
if sep_index == -1:
sep_index = raw.find('.')

# no separator -> treat everything as integer
if sep_index == -1:
digits_only = ''.join(normalize_digit(ch) for ch in raw)
if not digits_only:
return 0.0
value = float(digits_only)
return -value if is_negative else value

# split into integer and decimal part
int_raw = raw[:sep_index]
dec_raw = raw[sep_index + 1:]

# integer part: convert superscript/normal digits
integer_digits = ''.join(normalize_digit(ch) for ch in int_raw) or '0'

# decimal part: here we normally have the superscript digits
decimal_digits = ''.join(normalize_digit(ch) for ch in dec_raw) or '0'

number_str = f"{integer_digits}.{decimal_digits}"

try:
value = float(number_str)
except ValueError:
return 0.0

if is_negative:
value = -value

return value


def extract_amounts_from_period(text: str) -> list[float]:
"""
Extract amounts only from sentences of the form:
'din perioada ... în cuantum de <amount>'
and ignore sentences containing 'a fost acoperita suma de'
(meaning 'the amount was covered').
"""
if not text:
return []

pattern = re.compile(
r"din\s+perioada\s+([^\n\r]+?)\s*(?:\n|\r\n)?\s*(?:în|in)\s+cuantum\s+de?\s*"
r"([-−–]?\s*[0-9\s,.\u00B2\u00B3\u00B9⁰¹²³⁴⁵⁶⁷⁸⁹]+)",
flags=re.IGNORECASE
)

lower_text = text.lower()
amounts: list[float] = []

for match in pattern.finditer(text):
start = match.start()
amount_str = match.group(2)

# Skip contexts like "a fost acoperita suma de ..."
context_before = lower_text[max(0, start - 150):start]
if "a fost acoperita suma de" in context_before or "a fost acoperită suma de" in context_before:
continue

value = clean_and_convert_amount(amount_str)
amounts.append(value)

return amouts


After I collect the amounts, I format them and write them as text into Excel (using openpyxl) like this:

amounts = extract_amounts_from_period(text_from_pdf)
total = sum(amounts)

amounts_str = " + ".join(f"{v:.4f}".replace(".", ",") for v in amounts)
total_str = f"{total:.4f}".replace(".", ",")


Then, I put amounts_str and total_str into Excel cells

Problems

The decimals are not always converted correctly. Sometimes a value like 310,⁷⁶⁹⁷ becomes 310.0 or even 3107697 instead of 310.7697.

Negative amounts sometimes lose the minus sign. For example, −25,³⁴ in the PDF may end up as 25,34 (positive) in the Excel file.

Expected behavior

For the sample_text above I would expect:

extract_amounts_from_period(sample_text) == [310.7697, -25.34]


and later in Excel to see (as text):

310,7697 + -25,3400

Total cell: 285,4297

What is a robust way in Python to:

Parse these amounts from the PDF text (where decimals are written with superscript digits after a comma), and make sure the negative sign is always preserved, so that I get correct float values like 310.7697 and -25.34 before writing them to Excel?

Continue reading...