Research on Scene Text Recognition Algorithm Basedon Improved CRNN
Yilin Chen, Juan Yang · 2020
Image-based sequence recognition has always been a longstanding research topic in computer vision. Scene text recognition is one of the most important and challenging tasks in image-based sequence recognition. End-to-end scene text recognition based on deep learning now mainly transforms text recognition into sequence recognition problems. This paper improves the current advanced end-to-end trainable variable length recognition method CRNN [1]. Adding the IBN-Net [2] structure can improve the accuracy and generalization of the model without increasing the amount of calculation. Experiments in the public datasets ICDAR 2003 [3] and ICDAR 2013 [4] demonstrate the effectiveness of the proposed method. In addition, the proposed algorithm performs well in non-dictionary text recognition tasks, which clearly confirms its generalization.