PaddleOCR doesn't recognize digit 0

[Salmondx]

🔎 Search before asking

• I have searched the PaddleOCR Docs and found no similar bug report.
• I have searched the PaddleOCR Issues and found no similar bug report.
• I have searched the PaddleOCR Discussions and found no similar bug report.

🐛 Bug (问题描述)

PaddleOCR doesn't detect digit "0" in most cases that I encountered. I didn't find any settings that would help it to understand how this 0 looks like.

Attached one of the example images I tried to use it on to this issue:

In this sample image, PaddleOCR returns only 100 and 50 numbers, but not 0.

🏃‍♂️ Environment (运行环境)

OS Windows 11
Python 3.12.4
PaddleOCR 2.9.1
PaddlePaddle 2.6.2

🌰 Minimal Reproducible Example (最小可复现问题的Demo)

 paddleocr --image_dir image.png --lang en

[GreatV]

It seems like PaddleOCR is not recognizing the digit "0" in your images. Here are some possible reasons and solutions:

Possible Causes:

1. Font or Style Differences

   • If the digit "0" in your image has a unique font or style (e.g., a slashed zero or a rounded zero), the model may struggle to recognize it.

2. Pre-trained Model Limitation

   • The default PaddleOCR English model may not be optimized for all fonts or number styles.

3. Threshold or Confidence Issues

   • The model might detect the digit but with a low confidence score, causing it to be omitted in the final output.

4. Image Preprocessing Issues

   • If the contrast, background, or noise levels are affecting visibility, the OCR might fail to recognize certain characters.

Potential Solutions:

1. Try a Different Recognition Model

   • Instead of using the default English model, you can try a specialized digit recognition model provided by PaddleOCR:

     python3 tools/infer_rec.py -c configs/rec/rec_mv3_none_bilstm_ctc.yml \
       -o Global.pretrained_model=pretrain_models/v3_digital_mobile/best_accuracy \
       Global.infer_img="image.png"

   • This model is optimized for digit recognition and may perform better.

2. Check Character Dictionary

   • Ensure that "0" is included in the dictionary file used by the model. You can check or modify it in:

     ppocr/utils/dict/en_dict.txt
     

   • If "0" is missing, add it manually and retrain the model.

3. Use Image Preprocessing Techniques

   • Convert the image to grayscale and increase contrast:

     import cv2
     image = cv2.imread("image.png", cv2.IMREAD_GRAYSCALE)
     _, image = cv2.threshold(image, 128, 255, cv2.THRESH_BINARY)
     cv2.imwrite("processed_image.png", image)

   • Try running OCR on the processed image.

4. Train a Custom Model

   • If the issue persists, consider fine-tuning PaddleOCR with a dataset containing images of your specific font and style.

Related Discussion:

• A similar issue was discussed in this PaddleOCR Discussion where users faced recognition issues with digital-style numbers.

If none of these solutions work, you may need to train a custom model with labeled examples of your specific number style. Let me know if you need further assistance! 🚀

---

Response generated by feifei-bot | chatgpt-4o-latest