CHINESE-OCR
CHINESE-OCR is an open-source Python project for detecting and recognizing Chinese text in natural scene images. It uses TensorFlow for text detection and Keras or PyTorch for end-to-end OCR recognition of variable-length text. The system combines three neural networks: a VGG16-based classifier for detecting text orientation at 0, 90, 180, and 270 degrees, CTPN (CNN+RNN) for detecting horizontal text regions, and CRNN (CNN+GRU/LSTM+CTC) for recognizing the extracted text without pre-segmentation. CTC decoding eliminates the need for character-level position labels, making it effective for real-world scenarios where such annotations are unavailable. The project supports both GPU and CPU environments with Python 3.6 and TensorFlow 1.7, and provides pretrained model weights, training scripts, and datasets. Users can run demo.py with their own images to obtain text region detection and OCR results, or train the models on custom datasets. The implementation is primarily optimized for Chinese and English character