jonix6 / minepdf

Pure-Python PDF extraction tool based on PDFMiner

MinePDF

MinePDF is a pure-Python tool for extracting reflowable texts, images, fonts and other contents from PDF documents. This project is written based on PDFMiner, and performs a lot of improvements to make it easy for analyzing visual contents on pages.

This project is still under developing. Documentation, test suite and debugging will be completed later.

Improved features

• Variable parameters for grouping text lines into paragraphs
• Better image extraction using Pillow and Numpy
• Various color spaces support for color conversion
• Improved font parsing for glyph mapping and font extraction
• Improved render device for checking if objects visible
• Customizable extraction: page range selection and resource type filter
• Customizable logging system

Basic workflow

from minepdf import render_pages

fp = open('sample.pdf', 'rb')
pages = render_pages(fp, page_range='1-5,20-100', filtered=['path'])
for pageno, renderer, factory in pages: # extract each page
    handle_text(renderer.text_layer) # handle paragraphs
    handle_image(renderer.images) # handle images
    handle_resources(factory) # handle other resources (fonts, curves, etc.)
fp.close()

Text layout analysis

According to the LAParams in PDFMiner for layout analysis, this project provides multiple parameters for text layout analysis, even grouping text fragments into text lines and paragraphs.

Parameter	Definition
fontsize_between_ratio	minimum average font size ratio between both lines
fontsize_linebreak_ratio	minimum average font size ratio between the last character of the upper text line and the first character of the lower text line
max_line_spacing	maximum spacing between both text lines
max_word_spacing	maximum spacing between words in one line
indent_length	length of the indentation
min_column_gap	minimum column gap length among multi-column text blocks