Table Structure Recognition

Quick Start

To get started:

• Select a trained model from the Model to use dropdown
• Choose Output Type: HTML or JSON
• Configure header settings (Has Header and Header Type)
• Send table image path via msg.payload.image_path
• Provide detected table boxes via msg.payload.tables
• Provide OCR words via msg.payload.ocr
• Receive extracted table structure in msg.payload

Configuration

Model to use (required)

Select a pre-trained model from the dropdown menu. Models must be trained beforehand.

Output Type (required)

Choose between HTML or JSON output format.

• HTML: Returns table as HTML markup
• JSON: Returns structured JSON with cell data and coordinates

Has Header (required)

Specify if the table has header rows/columns. Options: True, False

When Has Header is True and Output Type is JSON, you must provide msg.payload.header_detail (and msg.payload.keyname_detail for dual-header layouts).

Header Type (required)

Specify header orientation. Options: horizontal_header, vertical_header.

Common Input Format

msg.payload.image_path (string)

Relative path of the table image on shared storage.

Example: "documents/table_image.png"

msg.payload.tables (array)

List of table bounding boxes in the image.

Format: [[x1, y1, x2, y2], ...]

msg.payload.ocr (array)

OCR text with bounding boxes.

Format: [[[x1, y1, x2, y2], "text"], ...]

msg.payload.header_detail (object, required for JSON with headers)

Header mapping used when output_type is json and has_header is True.

msg.payload.keyname_detail (object, required for some header layouts)

Key name mapping used for dual-header layouts.

Common Output Format

msg.payload contains an output field with the extracted table data.

HTML Output:

{"output": "<table><tr><th>Header</th></tr><tr><td>Data</td></tr></table>"}

JSON Output:

{"output": {"Header A": ["Cell 1", "Cell 2"], "Header B": ["Cell 3"]}}

Example

Input (msg.payload)

{
  "image_path": "documents/table_image.png",
  "tables": [[10, 20, 620, 380]],
  "ocr": [[[15, 25, 60, 45], "Header"], [[20, 60, 80, 90], "Value"]]
}

Output (msg.payload) - JSON

{
  "output": {
    "Header": ["Value"]
  }
}

Errors

When the block fails, it raises an error. Use a Catch block in your flow to handle failures and inspect the error payload.

Common mistakes

• Model not found: The selected model doesn't exist. Ensure the model is available.
• Invalid image path: The table image path is incorrect or file doesn't exist.
• No table detected: Image doesn't contain a recognizable table structure.
• Invalid output type: Output type must be either "html" or "json".
• Service unavailable: The service is unavailable or unreachable.

Best Practices

• Use clear, well-scanned table images for better extraction accuracy
• Ensure table borders are visible if possible
• Use JSON output when you need cell coordinates for further processing
• Use HTML output for direct display or simple data extraction
• Set Has Header correctly to improve extraction accuracy
• Use auto_detect for Header Type when header orientation is uncertain
• Test with sample tables to verify extraction quality
• Preprocess images (deskew, enhance contrast) for better results