opencv 14 OCR 分段字识别(segmented_word_recognition) vs2015

segmented_word_recognition使用tesseract，KNN，CNN三种算法。

如果要配置tesseract环境，参考opencv 13 OCR 端到端识别(end_to_end_recognition) vs2015；如果不配置tesseract环境，运行时，只有KNN，CNN两种结果。

01 资源

OpenCV自带的OCR分段字识别用例，opencv_contrib\modules\text\samples\segmented_word_recognition.cpp。segmented_word_recognition.cpp，可以识别一段文字。

模型文件(需要拷贝到Debug或Release目录中)：

opencv_contrib/modules/text/samples/OCRHMM_knn_model_data.xml.gz
opencv_contrib/modules/text/samples/OCRBeamSearch_CNN_model_data.xml.gz
opencv_contrib/modules/text/samples/OCRHMM_transitions_table.xml

测试图片：

opencv_contrib/modules/text/samples/scenetext_segmented_word01.jpg
opencv_contrib/modules/text/samples/scenetext_segmented_word01_mask.png
opencv_contrib/modules/text/samples/scenetext_segmented_word02.jpg
opencv_contrib/modules/text/samples/scenetext_segmented_word02_mask.png
opencv_contrib/modules/text/samples/scenetext_segmented_word03.jpg
opencv_contrib/modules/text/samples/scenetext_segmented_word03_mask.png
opencv_contrib/modules/text/samples/scenetext_segmented_word04.jpg
opencv_contrib/modules/text/samples/scenetext_segmented_word04_mask.png
opencv_contrib/modules/text/samples/scenetext_segmented_word05.jpg
opencv_contrib/modules/text/samples/scenetext_segmented_word05_mask.png

配置tesseract资源，参考opencv 13 OCR 端到端识别(end_to_end_recognition) vs2015。

02 编译segmented_word_recognition

参考opencv01 相对完整的编译opencv3.3.0 win版本，编译opencv3.3.0 vs2015版本。

生成文件：D:\git\opencv\build\v3.3.0\x64\bin\Debug\text-example-segmented_word_recognition.exe。

注：如果要运行tesseract结果，参考opencv 13 OCR 端到端识别(end_to_end_recognition) vs2015。

03 segmented_word_recognition项目配置

设置segmented_word_recognition项目为启动项。

# 如果路径中有空格，需要使用双引号,参数路径根据自己实际情况调整
配置属性==>调试==>命令参数=-@image=../../../../../../opencv_contrib/modules/text/samples/scenetext_segmented_word03.jpg -@mask=../../../../../../opencv_contrib/modules/text/samples/scenetext_segmented_word03_mask.png
配置属性==>调试==>工作目录=$(OutDir)

04 运行结果

scenetext_segmented_word03.jpg/scenetext_segmented_word03_mask.png结果：

Segmented word recognition.
A demo program on segmented word recognition. Shows the use of the OCRHMMDecoder API with the two provided default character classifiers.

Usage: text-example-segmented_word_recognition.exe [params] image mask

-?, -h, --help, --usage
print this message.
-l, --lex, --lexicon
(optional) lexicon provided as a list of comma separated words.

image (value:../../../../../../opencv_contrib/modules/text/samples/scenetext_segmented_word03.jpg)
source image for recognition.
mask (value:../../../../../../opencv_contrib/modules/text/samples/scenetext_segmented_word03_mask.png)
binary segmentation mask where each contour is a character.

OCR_Tesseract  output "Prlvate lee". Done in 132.917 ms.
OCR_NM         output "private Hire". Done in 167.645 ms.
OCR_CNN        output "Private Hire". Done in 6102.24 ms.