In practice, scene text recognition has been applied to various real-world scenarios, and it has become one of the most popular research topics in academic and industry for a long time. However, compared with the state-of-the-art scene text recognition algorithms, the text and digit recognition samples in OpenCV are out of date. These samples are based on traditional classifiers, which is difficult to handle the complex scenarios in the real world. Hence, I propose to improve the text recognition samples with the representative algorithms of scene text recognition from our research group. Specifically, CRNN is a good choice for its high accuracy and efficiency. Because there are some unsupported network layers and operations in OpenCV, I will modify the related parts of the original network. Moreover, it is necessary to accelerate the text recognizor with the CUDA support in OpenCV DNN Module. In conclusion, I propose to implement a fast text recognition sample in OpenCV, and build a complete text spotting pipeline for scene text detection and recognition.



Wenqing Zhang


  • Shiqi Yu