docs/en/useapis/ocrapi.md
::: tabs
== Baidu
https://ai.baidu.com/ai-doc/OCR/1k3h7y3db
https://ai.baidu.com/ai-doc/MT/dkifdqg54
https://fanyi-api.baidu.com/product/22
== Tencent
https://cloud.tencent.com/document/product/866/33526
https://cloud.tencent.com/document/product/551/17232
== Youdao
https://ai.youdao.com/doc.s#guide
== Volcano Engine
https://www.volcengine.com/docs/6790/116978
== Xunfei
https://www.xfyun.cn/doc/platform/quickguide.html
== Google Cloud Vision
https://cloud.google.com/vision/docs
== Ocr.space
== General Large Model Interface
Same as Translation
:::
The built-in PP-OCRv5_mobile (lightweight recognition model for Chinese, Japanese, and English) is already included. If you need to recognize other languages or want to use other models, you need to download the models in the settings and configure them for use.
The settings also provide some high-precision models, such as PP-OCRv6_medium and PP-OCRv5_server, which can achieve extremely high recognition accuracy, but the recognition speed is relatively slower.
| Model | Detection Module Hmean(%) | Recognition Module Avg Accuracy(%) | Supported Languages | Size(MB) |
|---|---|---|---|---|
| PP-OCRv6_small | 84.1 | 81.3 | Any | 25.2 |
| PP-OCRv6_medium | 86.2 | 83.2 | Any | 99.7 |
| PP-OCRv6_tiny | 80.6 | 73.5 | Any | 5.45 |
| PP-OCRv5_mobile | 79.0 | 81.29 | Simplified Chinese, Traditional Chinese, English, Japanese | 17.7 |
| PP-OCRv5_server | 83.8 | 86.38 | Simplified Chinese, Traditional Chinese, English, Japanese | 148 |
| eslav_PP-OCRv5_mobile | 79.0 | 81.6 | East Slavic languages | 11.2 |
| korean_PP-OCRv5_mobile | 79.0 | 88.0 | Korean | 12.2 |
| latin_PP-OCRv5_mobile | 79.0 | 84.7 | Latin-script languages | 11.3 |
To improve the recognition efficiency of high-precision models, the following methods can be adopted:
Use GPU Inference
If the software version being used is the Win10 version, or the system is Windows 11, the model can be set to run directly using the GPU.
Use OpenVINO Inference
If using an Intel CPU/NPU/GPU, the inference engine can be replaced with OpenVINO to accelerate recognition.
Download onnxruntime-openvino. After extracting, overwrite all files in runtimes/win-x64/native to LunaTranslator/files/DLL64, and then select the device to use.
::: tabs
== SnippingTool
[!WARNING] Only supports Windows 10 and Windows 11 operating systems.
If you are using the latest version of Windows 11, you can use it directly; otherwise, you need to install this module in Settings to use it.
== Manga-OCR
[!WARNING] This OCR engine performs poorly in recognizing horizontal text.
CPU Integration Package https://lunatranslator.org/Resource/IntegrationPack/manga_ocr/cpu
GPU Integration Package https://lunatranslator.org/Resource/IntegrationPack/manga_ocr/gpu
== WeChat/QQ OCR
Requires installation of WeChat or the latest version of QQ
== WindowsOCR
[!WARNING] Poor performance, not recommended for use.
[!WARNING] Only supports Windows 10 and Windows 11 operating systems.
https://learn.microsoft.com/en-us/windows/powertoys/text-extractor#supported-languages
== Tesseract5
[!WARNING] Poor performance, not recommended for use.
https://github.com/tesseract-ocr/tesseract/releases
:::