如何为 Textricator PDF OCR 阅读器设置 FSM 配置？

Posted 2023-04-17

【中文标题】如何为 Textricator PDF OCR 阅读器设置 FSM 配置？

【英文标题】：How to set the FSM configuaration for Textricator PDF OCR reader?

【发布时间】：2021-07-19 08:38:30

【问题描述】：

我正在尝试使用名为Textricator 的PDF 文档解析器。它可以使用 3 种不同的方法来解析带有一些常见 OCR 库的 PDF。 (itext5, itext7, pdfbox) 可用的方法有：text、table 和 form。 Text 用于普通的原始 OCR 识别，table 用于读取结构化表格数据，form 用于解析较少结构化的表单，使用 Finite状态机 (FSM)。

但是，我无法使用 form 解析器。也许我根本不明白如何组织许多配置状态。该文档缺少一个简单的表单示例，最近有人使用 form 方法发布了一个attempt to read a very basic table，但无法做到。我也试了一下，但没有成功。

问：谁能帮我在 YML 文件中配置状态机？ （这用于从该 repo 的 issues 之一解析演示文件，并显示在下面的复制屏幕截图中。）

---

---

YML 配置文件。

extractor: "pdf.pdfbox"

header:
  default: 100
footer:
   default: 600

maxRowDistance: 2

rootRecordType: item
recordTypes:
  item:
    label: "item"
    valueTypes:
      - item
      - date
      - description
      - order_number
      - quantity
      - price

valueTypes:
  item:
    label: "Item"
  date:
    label: "Date"
  description:
    label: "Description"
  order_number:
    label: "OrderNo"
  quantity:
    label: "Qty"
  price:
    label: "Price"
 
initialState: "INIT"

states:
  INIT:
    transitions:
      -
        condition: item
        nextState: item

  item:
    startRecord: true
    transitions:
      -
        condition: date
        nextState: date  

  date:
    include: true
    transitions:
      -
        condition: description
        nextState: description  

  description:
    include: true
    transitions:
      -
        condition: description
        nextState: description     
      -
        condition: order_number
        nextState: order_number
      -
        condition: quantity
        nextState: quantity

  order_number:
    include: true
    transitions:
      -
        condition: order_number
        nextState: order_number
      -
        condition: quantity
        nextState: quantity

  quantity:
    include: true
    transitions:
      -
        condition: price
        nextState: price

  price:
    include: true
    transitions:
      -
        condition: end
        nextState: end

  end:
    include: false
    transitions:
      -
        condition: any
        nextState: end

conditions:

  item:         '73 < ulx < 110 and text =~ /(\\d)*/'
  date:         '110 < ulx < 181 and text =~ /([0-9\-]*)/'
  description:  '193 < ulx < 366'
#  order_number: '12 <= uly_rel <= 16 and text =~ ^.+/((\d6)\-)((\d2))/'
  order_number: '12 <= uly_rel <= 16 and text =~ ^.+((\d6)\-)((\d2))'
  quantity:     '393 < ulx < 459'
  price:        '459 < ulx < 523'

  end:          'text =~ /(Footer)/'
  any: "1 = 1"

您可能想知道为什么我在这个简单的示例中坚持使用 form 处理器，但这是因为在我的实际文档中，我将有一个更复杂的子项目子结构描述字段。这只能（？）由状态机 AFAIK 有效处理。

但是，也许这不是适合这项工作的工具？那么还有哪些其他选择呢？

---

更新：（2021-05-18）

Textricate 的作者现在已经修改了使用的库、文档并更正了几个工作示例和用户问题。感谢用户 mweber，我现在有了一个完美运行的解析器，不再需要使用 awk 到handle weird columns。

【参考方案1】：

由于 Textricator 是一种用于解析 imo 的 pdf 的隐藏宝石，我很高兴看到有人使用它并将使用示例文档的配置发布到 github 问题：

extractor: "pdf.pdfbox"

header:
  default: 100
footer:
  default: 600

maxRowDistance: 2

rootRecordType: item
recordTypes:
  item:
    label: "item"
    valueTypes:
      - item
      - date
      - description
      - order_number
      - quantity
      - price

valueTypes:
  item:
    label: "Item"
  date:
    label: "Date"
  description:
    label: "Description"
  order_number:
    label: "OrderNo"
  quantity:
    label: "Qty"
  price:
    label: "Price"

initialState: "INIT"

states:
  INIT:
    include: false
    transitions:
      -
        condition: item
        nextState: item
      - condition: any
        nextState: INIT

  item:
    startRecord: true
    transitions:
      -
        condition: date
        nextState: date  

  date:
    include: true
    transitions:
      -
        condition: description
        nextState: description  

  description:
    include: true
    transitions:
      -
        condition: description
        nextState: description     
      -
        condition: order_number
        nextState: order_number
      -
        condition: quantity
        nextState: quantity
      -
        condition: item
        nextState: item

  order_number:
    include: true
    transitions:
      -
        condition: order_number
        nextState: order_number
      -
        condition: quantity
        nextState: quantity

  quantity:
    include: true
    transitions:
      - 
        condition: price
        nextState: price

  price:
    include: true
    transitions:
      -
        condition: end
        nextState: end
      - 
        condition: description
        nextState: description
      -
        condition: item
        nextState: item

  end:
    include: false
    transitions:
      -
        condition: any
        nextState: end

conditions:

  item:         '73 < ulx < 110 and text =~ /(\\d)*/'
  date:         '110 < ulx < 181 and text =~ /([0-9\\-]*)/'
  description:  '193 < ulx < 366'
  order_number: '12 <= uly_rel <= 16 and text =~ /^.+(([0-9]6)\\-)(([0-9]2))/'
  quantity:     '393 < ulx < 459'
  price:        '459 < ulx < 523'

  end:          'text =~ /(Footer)/'
  any: "1 = 1"

【讨论】：

太棒了！ SW 作者还在 recordTypes 下添加了filter: 'quantity > 0'（带有 4 个空格缩进），在 valueTypes 的 quantity 下添加了type: number。