OCR

Overview

OCR node detects text regions in an input frame and recognizes the characters within those regions.

Use this node to extract printed or handwritten text from documents, labels, signs, or displays. The node runs a two-stage pipeline: a detection model first locates text bounding boxes, then a recognition model reads the characters inside each box.

Input

Input Image

image required

The image frame to process. Connect this to a camera or upstream image output.

Detection model

string required

Model used to locate text regions in the frame.

Values:

• PPOCRv6_DET_SMALL — PP-OCRv6 Small detector.
• PPOCRv6_DET_TINY — PP-OCRv6 Tiny detector; faster, lighter.
• DB (default) — Differentiable Binarization; accurate for most printed text scenarios.
• NONE — skips detection and passes the entire frame to the recognition model directly.

Recognition model

string required

Model used to read characters from each detected text region.

Values:

• PPOCRv6_REC_SMALL — PP-OCRv6 Small recognizer.
• PPOCRv6_REC_TINY — PP-OCRv6 Tiny recognizer.
• PPOCRv5_REC_MOBILE_TH — PP-OCRv5 Mobile Thai recognizer.
• CRNN (default) — Convolutional Recurrent Neural Network; accurate for printed text.

Language

string required

Recognition language for PP-OCRv6 recognizers (PPOCRv6_REC_SMALL / PPOCRv6_REC_TINY). Appears only for those recognition models.

Default: en

Common values include en (English), ch (Chinese Simplified), chinese_cht (Chinese Traditional), japan (Japanese), and many other PP-OCR language codes.

Overlay results

boolean required advanced

Whether to draw text region boxes and recognized text on the output frame. See Overlay Results.

Default: true

Draw lines

boolean required advanced

Draw text-region outlines on the overlay.

Default: true

Draw text

boolean required advanced

Draw recognized text strings on the overlay.

Default: true

Draw confidence

boolean advanced

Draw recognition confidence scores on the overlay.

Default: false

Confidence threshold

number advanced

Minimum detection confidence for keeping a text region. Shown when Detection model is not NONE.

Range: 0.0 to 1.0
Default: 0.5

Box Width Range / Box height Range

array advanced

Pixel [min, max] filters for detected text box width and height. Shown when Detection model is not NONE. Boxes outside the ranges are discarded.

Default: [0, 100000] for each

Detection pixel threshold / Detection box threshold / Unclip ratio

number advanced

PP-OCRv6 detection tuning (when Detection model is PPOCRv6_DET_SMALL or PPOCRv6_DET_TINY):

• Detection pixel threshold — pixels above this score are treated as text (default 0.3).
• Detection box threshold — boxes whose average pixel score is above this are kept (default 0.6).
• Unclip ratio — expansion coefficient for detected text regions (default 1.5).

Binary threshold / Polygon threshold / Unclip ratio / Maximum result / Padding Scale

number integer array advanced

DB-specific detection tuning (when Detection model is DB):

• Binary threshold (default 0.3)
• Polygon threshold (default 0.5)
• Unclip ratio (default 1.5)
• Maximum result (default 100)
• Padding Scale — [width, height] scale factors (default [1, 1])

Recognition score threshold

number advanced

Minimum recognition score to keep a text result. Shown for PP-OCR recognizers.

Range: 0.0 to 1.0
Default: 0

Text line orientation

boolean advanced

Classify and correct text-line orientation before recognition (PP-OCR recognizers).

Default: false

Use TensorRT if available

boolean advanced

Try TensorRT on Nvidia GPUs when using PP-OCR detection or recognition models. The first run may convert the model; later runs are faster.

Default: true

Output

Overlay Image

image

Output frame from the node. If overlays are enabled, text regions and recognized strings are annotated on this frame.

Detected Count

integer

Number of recognized text regions in the current frame.

Detected Texts

array

Array of recognized text objects. Each object contains:

• Points (contour) — polygon points of the text region.
• Bounding Box (bbox) — [x, y, width, height].
• Detected Text (label) — the recognized string.