Service Configuration: Extract Text
The Extract Text section in Service Profile configures dedicated text-extraction models. Models in Generic AI can also provide Extract Text when they support vision input and have that capability enabled.
What the Extract Text service does
Section titled “What the Extract Text service does”Extract Text reads characters inside existing text blocks and writes the recognition result into each block’s source text. It is separate from Detect Text Block:
- Detection finds text regions and creates text blocks.
- Extraction recognizes the characters in those regions. Some detection services also return text, so a separate extraction step may not be needed.
Currently supported service
Section titled “Currently supported service”The dedicated Extract Text service currently supports only the Local Text Extractor.
It uses a local PP-OCRv6 model. No online OCR API is required, and recognition data is not sent over the network for remote processing.
Download or import the model files in the app’s Service Profile/local model management area before running extraction. Extraction is unavailable until the model is installed.
Technology and workflow
Section titled “Technology and workflow”The local extractor uses a PP-OCRv6 recognition model. Its workflow is roughly:
- Text blocks must already exist. Usually, run text-block detection first or create the blocks manually.
- The app crops the image region for each text block and sends that crop to the local recognition model.
- The model returns text lines or columns and their positions. The app uses layout algorithms to combine them into a single text string, filter out some irrelevant lines or columns, and infer text orientation.
- The recognized text is saved as the text block’s source text for later translation and review.
The model files must be obtained before use. They can be downloaded or imported offline through the local model management interface.