OCR (Optical Character Recognition)

Quick Start

To get started:

• Choose a model from the Model to use dropdown
• Select OCR Type (words, lines, handwritten, etc.)
• Send image path via msg.payload.image_path
• Optionally specify regions via msg.payload.regions
• Receive OCR results in msg.payload

Configuration

Model to use (required)

Select an OCR model from the dropdown menu.

OCR Type (required)

Type of OCR to perform:

• words_and_bounding_boxes: Extract words with coordinates
• line_detection_and_recognition: Extract text lines
• line_detection_and_recognition_with_handwritten_text: Handle mixed handwritten/printed text
• complete handwritten OCR: For fully handwritten documents
• partial handwritten OCR: For documents with some handwritten content

Box Threshold (optional)

Confidence threshold (0-1) for text detection boxes. Default: 0.3

Text Threshold (optional)

Confidence threshold (0-1) for text recognition. Default: 0.3

Common Input Format (All Algorithms)

msg.payload.image_path (string)

Relative path of the image file on shared storage.

Example: "documents/scan.jpg"

msg.payload.regions (array, optional)

Specific regions to perform OCR on.

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

Output by OCR Type

msg.payload contains an output field with the OCR results.

words_and_bounding_boxes / complete handwritten OCR / partial handwritten OCR

msg.payload.output is a list of entries in the format: [[bbox, text, confidence], ...]

bbox is either [x1, y1, x2, y2] or a polygon [x1, y1, x2, y2, x3, y3, x4, y4].

Example: {"output": [[[100, 50, 200, 80], "Invoice", 98], [[300, 200, 400, 230], "Total", 92]]}

line_detection_and_recognition / line_detection_and_recognition_with_handwritten_text

msg.payload.output is a list of line entries in the format: [[line_bbox, words_and_bboxes, line_text], ...]

words_and_bboxes uses the same [bbox, text, confidence] entries described above.

Example: {"output": [[[90, 40, 420, 90], [[[100, 50, 200, 80], "Invoice", 98]], "Invoice"]]}

Region-based Outputs

If region selection is enabled in configuration:

• For coordinates-based selection, msg.payload.output is an object like { "1": { "region": [x1, y1, x2, y2], "ocr": [...] }, ... }
• For other region types, msg.payload.output is an object like { "0": { "region": [x1, y1, x2, y2], "output": [...] }, ... }

Example

Input (msg.payload)

{
  "image_path": "documents/scan.jpg",
  "regions": [[0, 0, 1200, 800]]
}

Output (msg.payload)

{
  "output": [
    [[100, 50, 200, 80], "Invoice", 98],
    [[300, 200, 400, 230], "Total", 92]
  ]
}

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

• Missing image path: msg.payload.image_path is required and must point to a file on shared storage.
• Invalid regions format: If provided, msg.payload.regions must be [[x1, y1, x2, y2], ...].
• Threshold out of range: Threshold values must be between 0 and 1.

Best Practices

• Use clear, high-resolution images for better OCR accuracy
• Choose appropriate OCR type based on your document content
• Use region-based OCR when only specific areas need text extraction
• Adjust thresholds to balance between recall and precision
• Preprocess images (deskew, enhance contrast) for better results
• Use handwritten-specific models for handwritten content
• Always validate OCR results in critical applications