tesseract-ocr

OCR : 光学字符识别(Optical character recognition)，是将打字，手写或印刷的文本图像转换成计算机文本。输入可以是扫描的文档，文档的照片，还是场景的照片（身份证照片、驾驶证、广告牌）。

Tesseract：开源的OCR引擎，在Apache 2.0协议下可用。

安装

下载地址，我当前下载的版本是tesseract-ocr-w32-setup-v4.1.0.20190314.exe，按照步骤安装就行。我的安装路径是E:\Tesseract-OCR。

用命令行工具测试一下

1
2
3
4
$ tesseract tesseract-ocr-1.png tesseract-ocr-1.output
Tesseract Open Source OCR Engine v4.0.0.20190314 with Leptonica
Warning: Invalid resolution 0 dpi. Using 70 instead.
Estimating resolution as 108

我输入的图片是

输出的结果 tesseract-ocr-1.output.txt中的文本是

1
2
tesseract-ocr
Hello World!

可以直接输出到stdout

1
2
3
4
5
6
$ tesseract tesseract-ocr-1.png stdout
tesseract-ocr
Hello World!

Warning: Invalid resolution 0 dpi. Using 70 instead.
Estimating resolution as 108

可以添加-l选项指定语言，如果不指定，默认是eng.

默认安装后，语言列表仅有eng和osd。

1
2
3
4
$ tesseract --list-langs
List of available languages (2):
eng
osd

训练

我们有如下一张图片，进行识别

1
2
3
4
5
$ tesseract tesseract-ocr-2.png stdout
40293847 S565647386291L0

Warning: Invalid resolution 0 dpi. Using 70 instead.
Estimating resolution as 147

识别的结果显然不是我们想要的，接下来我们对它进行训练。

  1. 使用在线工具将jpg转换成tiff格式，重命名为l2m2hw.normal.exp0.tif。

    tiff命名规则：${lang}.${fontname}.exp${num}

  2. 生成box文件

    1
    2
    3
    4
    $ tesseract --psm 6 --oem 3 l2m2hw.normal.exp0.tif l2m2hw.normal.exp0 makebox
    Tesseract Open Source OCR Engine v4.0.0.20190314 with Leptonica
    Page 1
    Warning: Invalid resolution 0 dpi. Using 70 instead.

    通过帮助命令可以查看psm和oem参数的可选项及含义

    1
    2
    3
    4
    5
    6
    7
    8
    9
    10
    11
    12
    13
    14
    15
    16
    17
    18
    19
    20
    21
    22
    23
    $ tesseract --help-psm
    Page segmentation modes:
      0    Orientation and script detection (OSD) only.
      1    Automatic page segmentation with OSD.
      2    Automatic page segmentation, but no OSD, or OCR. (not implemented)
      3    Fully automatic page segmentation, but no OSD. (Default)
      4    Assume a single column of text of variable sizes.
      5    Assume a single uniform block of vertically aligned text.
      6    Assume a single uniform block of text.
      7    Treat the image as a single text line.
      8    Treat the image as a single word.
      9    Treat the image as a single word in a circle.
     10    Treat the image as a single character.
     11    Sparse text. Find as much text as possible in no particular order.
     12    Sparse text with OSD.
     13    Raw line. Treat the image as a single text line,
           bypassing hacks that are Tesseract-specific.
    $ tesseract --help-oem
    OCR Engine modes:
      0    Legacy engine only.
      1    Neural nets LSTM engine only.
      2    Legacy + LSTM engines.
      3    Default, based on what is available.

  3. 打开jTessBoxEditor (运行E:\jTessBoxEditor\training.bat）, Box Editor->Open, 选择刚才生成的tif文件。

    jTessBoxEditor是训练Tesseract的辅助工具。下载地址。我当前下载的版本是2.3.1。

  4. 在左侧修正字符后，保存

  5. 创建font_properties文件

    1
    $ echo normal 0 0 0 0 0> font_properties

    语法：<fontname> <italic> <bold> <fixed> <serif> <fraktur>

    到这一步，你应该有这些文件：

    1
    2
    $ ls
    font_properties  l2m2hw.normal.exp0.box  l2m2hw.normal.exp0.tif

  6. 生成训练文件 .tr

    1
    2
    3
    4
    5
    6
    7
    8
    9
    $ tesseract l2m2hw.normal.exp0.tif l2m2hw.normal.exp0 nobatch box.train
    Tesseract Open Source OCR Engine v4.0.0.20190314 with Leptonica
    Page 1
    Warning: Invalid resolution 0 dpi. Using 70 instead.
    Estimating resolution as 147
    APPLY_BOXES:
       Boxes read from boxfile:      20
       Found 20 good blobs.
    Generated training data for 2 words

  7. 生成unicharset字符集文件

    1
    2
    3
    $ unicharset_extractor l2m2hw.normal.exp0.box
    Extracting unicharset from box file l2m2hw.normal.exp0.box
    Wrote unicharset file unicharset

  8. 生成特征字符文件

    1
    2
    3
    4
    5
    6
    $ mftraining -F font_properties -U unicharset -O l2m2hw.unichaset l2m2hw.normal.exp0.tr
    Done!
    Read shape table shapetable of 10 shapes
    Reading l2m2hw.normal.exp0.tr ...
    Warning: no protos/configs for Joined in CreateIntTemplates()
    Warning: no protos/configs for |Broken|0|1 in CreateIntTemplates()

  9. 生成字符正常化特征文件 normproto

    1
    2
    3
    4
    5
    $ cntraining l2m2hw.normal.exp0.tr
    Reading l2m2hw.normal.exp0.tr ...
    Clustering ...

    Writing normproto ...

    到这一步，你有这些文件了。

    1
    2
    3
    4
    $ ls
    font_properties         l2m2hw.normal.exp0.tif  normproto   unicharset
    inttemp                 l2m2hw.normal.exp0.tr   pffmtable
    l2m2hw.normal.exp0.box  l2m2hw.unichaset        shapetable

  10. 重命名文件

    1
    2
    3
    4
    5
    $ mv inttemp l2m2hw.inttemp
    $ mv normproto l2m2hw.normproto
    $ mv pffmtable l2m2hw.pffmtable
    $ mv shapetable l2m2hw.shapetable
    $ mv unicharset l2m2hw.unicharset

  11. 合并训练文件

    1
    2
    3
    4
    5
    6
    7
    8
    9
    10
    $ combine_tessdata l2m2hw.
    Combining tessdata files
    Output l2m2hw.traineddata created successfully.
    Version string:v4.0.0.20190314
    1:unicharset:size=721, offset=192
    3:inttemp:size=134520, offset=913
    4:pffmtable:size=110, offset=135433
    5:normproto:size=1382, offset=135543
    13:shapetable:size=184, offset=136925
    23:version:size=15, offset=137109

  1. 测试l2m2hw训练库

    拷贝训练库到安装目录的tessdata下

    1
    $ cp l2m2hw.traineddata /e/Tesseract-OCR/tessdata/

    再次识别

    1
    2
    3
    4
    5
    $ tesseract tesseract-ocr-2.png stdout -l l2m2hw
    10293847 565647382910

    Warning: Invalid resolution 0 dpi. Using 70 instead.
    Estimating resolution as 147

    正确啦！

Reference