Extract tables from pdf using pdfplumber and pytesseract

pdfplumber-tesseract.py

# Extracting tabular data from pdf using Python pdfplumber together with Tesseract OCR
# Author Jarkko Saltiola 2021 (MIT License, Python 3.8.6)

# Pdfplumber, tabula, camelot and probably some other PDF parser utilities have hard 
# time parsing tables that have column data overlapping over other columns, and
# probably on many other cases too.

# Pdfplumber gives a good level of control for splitting pdf into parts which can
# be read with it's methods or be passed for pytesseract as PIL image.

# This can be a starting point for hooking these two together.
# I abandoned this as it started getting too complex and there was a better workaround. 
# Main issue was matching the row values together and occasional unmatching count of rows that both tools returned.

# Could be improved by matching row values with some intelligence. 
# Or row positions could be hard coded so their count would stay the same.
# And maybe run tesseract for all the pre-cropped cells.

import pdfplumber
import pytesseract

ts = {
    "vertical_strategy": "text",  # "lines" is the default option that didn't work
    "horizontal_strategy": "text",
    "keep_blank_chars": False,  # Try with True as well if False doesn't give good results
}

pdf = pdfplumber.open("inventory.pdf")

for page in pdf.pages:
    # Extract first (EAN) column with pdfplumber table extractor

    # Select crop box eg. with MacOS Preview, rectangular crop tool, info window crop area.
    crop_ean = page.crop([32,52,110,780])
    # crop_ean_img = crop_ean.to_image()
    # crop_ean_img.original.show()  # preview the crop result
    crop_table_ean = crop_ean.extract_table(table_settings=ts)


    # Extract the third (count) column with Tesseract OCR

    crop_count = page.crop([350,71,410,71+710])
    # Pass cropped image to pytesseract, use higher resolution (default 72) for better accuracy.
    crop_count_img = crop_count.to_image(resolution=300)
    # crop_count_img.original.show()

    # Tesseract setting to read columns and only numbers
    ocr = pytesseract.image_to_string(crop_count_img.original,config='--psm 6 -c tessedit_char_whitelist=0123456789')

    ocr_list = ocr.split("\n")  # string to list

    if ocr_list[-1] == '\x0c':  # clean carbage item from the end
        ocr_list = ocr_list[:-1]


    print(f"{len(crop_table_ean)} {len(ocr_list)}")  # These counts should match reliably