Ainized-CRAFT-pytorch (ENG)
Scene text detection is the process of detecting the text present in the image, followed by surrounding it with a rectangular bounding box. Since performances of camera and smart phone has improved, scene text detection became important part of application such as extracting text from license plate image and so on.
With the development of deep learning, scene text detection methods based on neural networks have emerged. However, most of the existing datasets of the scene text detection provide word-level annotations, so the methods trained with a rigid word-level bounding box. This method allows for recognition of horizontal words, but there is a disadvantage that they do not recognize words that are curved, arbitrarily-oriented, or deformed.
To overcome these disadvantages, CRAFT(Character Region Awareness for Text Detection) recognizes each character and is said to group the characters together. CRAFT produce two things using CNN:
region score - localize individual characters in the image
affinity score - group each character into a single instance
To produce character regions, dataset of character-level annotation is required. But as described above, because most of the existing datasets provide word-level annotations, there is a lack of character-level annotations. To solve this problem, CRAFT generate character boxes from word-level annotation in a weakly supervised manner.
The image below is the result of performing the above picture through CRAFT.
You can test it through the following button.
Reference
https://data-newbie.tistory.com/187
https://github.com/clovaai/CRAFT-pytorch
https://arxiv.org/abs/1904.01941
https://www.mytoys.de/pattex-pattex-crocodile-power-klebeband-silber-30-m-13947723.html