Xiangyun Open Platform Document and Table Recognition

Problem Addressed

Documents such as scanned tables, invoices, and forms often exist as PDF, OFD, or image files, making manual transcription slow and error-prone. Xiangyun Open Platform Document and Table Recognition is useful for batch OCR of unstructured documents, producing processable text for downstream parsing, storage, or table extraction.

How the Skill Works

The skill relies on Xiangyun OCR credentials netocr_key and netocr_secret, which must be configured before first use. A typical flow is:
- Prepare input files: supports .pdf, .ofd, and image formats such as .jpg, .png, .bmp, .tif, .webp;
- Select a language code: for example, Simplified Chinese printed text uses 0, English uses 2, Japanese uses 11, and handwritten options map to other codes;
- Run batch recognition: files are sent to Xiangyun OCR, billed per call, and returned for further processing.

It fits engineering workflows that need multilingual documents normalized into text, especially scanned PDF/OFD files, receipts, and table screenshots.

Boundaries and Notes

The materials do not promise fully structured table output. If the goal is direct row and column extraction, treat OCR text as an intermediate result and add table parsing logic. Images should be clear, preferably larger than 500px in width and height; single files should stay under 10MB. The config file is stored as config.json in the skill directory. Credentials are sensitive and should not be committed to repositories or written to logs. Xiangyun OCR is billed per call, so validate cost and recognition quality on a small sample before running large batches.