CRNN+CTC OCR Relative

从自然场景图片中进行文字识别，需要包括2个步骤：文字检测：解决的问题是哪里有文字，文字的范围有多少;文字识别：对定位好的文字区域进行识别，主要解决的问题是每个文字是什么，将图像中的文字区域进转化为字符信息。

1. DAR&OCR&STR

• Document Analysis& Recognition: aims at the automatic extraction information presented on paper.
  • Documnet Image Enhancement; Layout Analysis; Character Recognition;
• Optical Character Recognition:
  • conversion of images of handwritten or printed text into machine-encoded text
  • the image may come from a scanned document, photo of a document, a scene-photo
• Scene Text Recognition: recognizing text in natural scene.

• https://mp.weixin.qq.com/s/F1d_pZQoVeUd9Uy5Z0Hc1Q

1. 基于RNN文字识别框架

1. CNN+RNN+CTC(CRNN+CTC)
2. CNN+Seq2Seq+Attention

From: https://zhuanlan.zhihu.com/p/43534801