pdf2txt_mnbvc: A Python repository from jayhenry

[TOC]

Intro

将pdf转换为txt文件，并且检查其分栏排版，以及是否有图像、表格。

Usage

见 examples/example.py

import pdf2txt

src_file = '../data/逍遥游.pdf'
dest_file = './逍遥游.txt'

pdf2txt.convert(src_file, dest_file)

Features

关于换行符和换页符

有多个字符表示类似含义，在不同系统上默认的方式不同，但是现代文本编辑器一般都兼容这些方式。

Unix和OS X使用LF(也称为NL)作为换行符。LF在编程语言里通常表示为转义符\n，他的ASCII值为10或者0xA。

老的Mac上使用CR作为换行符。CR通常表示为转义符\r，他的ASCII值为13或者0xD。

Windows使用CR+LF作为换行符。他是前面两个字符串联使用\r\n。

还要一个特殊字符叫换页符（Form Feed），通常用来表示换到下一页，但是现在用的不多。

https://stackoverflow.com/questions/3091524/what-are-carriage-return-linefeed-and-form-feed

CR: Carriage return means to return to the beginning of the current line without advancing downward. The name comes from a printer's carriage, as monitors were rare when the name was coined. This is commonly escaped as "\r", abbreviated CR, and has ASCII value 13 or 0xD.

LF or NL: Linefeed means to advance downward to the next line; however, it has been repurposed and renamed. Used as "newline", it terminates lines (commonly confused with separating lines). This is commonly escaped as "\n", abbreviated LF or NL, and has ASCII value 10 or 0xA. CRLF (but not CRNL) is used for the pair "\r\n".

FF: Form feed means advance downward to the next "page". It was commonly used as page separators, but now is also used as section separators. Text editors can use this character when you "insert a page break". This is commonly escaped as "\f", abbreviated FF, and has ASCII value 12 or 0xC.

As control characters, they may be interpreted in various ways.

The most important interpretation is how these characters delimit lines. Lines end with NL on Unix (including OS X), CRLF on Windows, and CR on older Macs. Note the shift in meaning from LF to NL, for the exact same character, gives the differences between Windows and Unix, which is also why many Windows programs use CRLF to separate instead of terminate lines. Many text editors can read files in any of these three formats and convert between them, but not all utilities can.

Form feed is much less commonly used. As page separator, it can only come between lines or at the start or end of the file.

More about pdf & PyMuPDF

PyMuPDF Basic Intro

https://pymupdf.readthedocs.io/en/latest/the-basics.html#extract-text-from-a-pdf

PyMuPDF Page

https://pymupdf.readthedocs.io/en/latest/app1.html

TextPage is one of (Py-) MuPDF’s classes. It is normally created (and destroyed again) behind the curtain, when Page text extraction methods are used, but it is also available directly and can be used as a persistent object. Other than its name suggests, images may optionally also be part of a text page:

<page>
    <text block>
        <line>
            <span>
                <char>
    <image block>
        <img>

A text page consists of blocks (= roughly paragraphs).

A block consists of either lines and their characters, or an image.

A line consists of spans.

A span consists of adjacent characters with identical font properties: name, size, flags and color.

TextPage结构见下图，来自 https://pymupdf.readthedocs.io/en/latest/textpage.html#structure-of-dictionary-outputs

其他材料 https://pymupdf.readthedocs.io/en/latest/tutorial.html#working-with-pages

https://pymupdf.readthedocs.io/en/latest/page.html

页面坐标系

https://pymupdf.readthedocs.io/en/latest/rect.html#Rect.round

页面坐标单位&fonesize单位

按自然阅读顺序提取文本

最简单的提取方法，会按pdf文件添加元素的顺序进行提取。
而有时候为了防止copy，一些pdf会打乱添加元素的顺序，但是排版上不影响阅读。这样简单地抽取文本就无法按照正常的阅读顺序排列，见这个例子。
- 为了解决上面的情况，有一些办法，一个简单的方法是讲文本Block根据坐标位置排序，按从上到下从左到右的顺序，像这个例子
另外，有一个复杂一些的方法，这个代码中的page_layout函数考虑到排版格式，大概流程是
- 得到每个字符的位置信息，并且去掉旋转的字符
- 根据字符的位置，计算所有行坐标，忽略间隔小于GRID的行坐标
- 遍历每一行，得到每行从左到右的字符列表
- 计算字符宽度slot，用于后续判断空格
- 根据上面的信息生成每行的string
- 需要注意，通过此方法生成的text保持了原来pdf的排版信息。如pdf是三栏，输出的text中也是三栏，并不是文字流。

这些信息来源于 https://github.com/pymupdf/PyMuPDF-Utilities/blob/master/text-extraction/README.md.

提取表格

没有简单的方法能准确地判断表格在页面中的位置，这通常是一个需要AI、ML技术来解决的复杂问题。但是对于一些简单的情况，可以使用PyMuPDF提供的vector graphics分析工具来判断表格的存在，比如是否有直线、矩形框等。这些信息来源于 https://github.com/pymupdf/PyMuPDF-Utilities/blob/master/table-analysis/README.md.

https://github.com/pymupdf/PyMuPDF-Utilities/blob/master/examples/extract-table/README.md

什么是PDF表单(PDF Form)？

https://helpx.adobe.com/cn/acrobat/using/create-form.chromeless.html

jayhenry/pdf2txt_mnbvc