Information Reading (OCR)
This page describes the configuration workflow for OCR-based information reading. It is used to recognize printed, engraved, or inkjet characters on target objects and extract key information such as model and batch.
Click Configuration Wizard, select Information Reading, then choose Optical Character Recognition.
Workflow
The complete workflow includes four stages:
-
Image Preprocessing
-
Pose Alignment
-
Information Reading
-
General Settings
Image Preprocessing
Before recognition, you can enable Convert Image Color Space or Image Preprocessing to improve target features.
Convert Image Color Space
Convert input image from one color space to another (for example, BGR to Gray or BGR to HSV) to highlight features for subsequent processing.
For details, see Convert Image Color Space.
Image Preprocessing Parameters
Supports enhancement, denoising, morphology, grayscale inversion, and edge extraction.
For details, see Image Preprocessing.
Pose Alignment
After preprocessing, configure pose alignment so target pose in current image is corrected to match template pose.
Add Alignment Settings
Create a parameter group for pose alignment. Multiple groups are supported and independent from each other.
Click Add to create a new group, choose alignment mode, and configure parameters.
Supported modes:
-
No Alignment: Use input image directly without pose correction.
-
2D Alignment: Align through translation/rotation with edge-based matching. See 2D Alignment.
-
2D Blob Alignment: Align based on selected Blob centroid and principal axis. See 2D Blob Alignment.
After creating a group, right-click group name (or use action button) to rename, delete, or duplicate.
2D Alignment
2D Alignment uses translation and rotation to align target object in input image to template.
Set Recognition Region
Set effective alignment area. Region should fully cover target object with proper margin.
-
Whole Image as Recognition Region: Use entire image.
-
Custom Recognition Region: Manually draw region and ignore unrelated background.
Recognize Target Object
Configure Target Template
After region setup, choose/edit template in 2D template editor by clicking Edit.
Select representative and stable edge features to ensure unique and accurate matching. For details, see 2D Matching Template Editor.
| Click Update after each template edit. |
Adjust Recognition Parameters
Click Run Step to view matching result and tune parameters if needed.
For details, see 2D Alignment.
Click Next to continue.
2D Blob Alignment
2D Blob Alignment detects blobs, selects target Blob by geometric features, then aligns centroid and principal axis.
Set Recognition Region
Set effective area with sufficient margin. Rectangle and circle region modes are supported, and multiple regions can be mixed.
Recognize Target Object
Tune parameters according to target features.
For details and tuning examples, see 2D Blob Alignment.
Information Reading
After alignment, this stage recognizes letters, digits, and symbols through deep-learning OCR.
Capture Images
-
Ensure image input is connected.
-
One image is captured automatically when entering the tool. Click Capture Image for additional samples.
|
Collect diverse samples that cover position/angle, lighting/background, and appearance changes. |
Set Target Region
-
Click Edit to enter ROI configuration.
-
Set single-character size by adjusting the orange character-size box.
-
Set target region:
-
Whole image as target region
-
Custom target region (rectangle or annulus)
-
If annulus is used, set reading direction as clockwise or counterclockwise.
-
(Optional) Configure masked area using polygon selection.
-
Click Save and Use.
Model Validation and Optimization
After ROI setup, click Validate.
Set Validation Parameters and Verify Effect
| Parameter | Description |
|---|---|
Validation Result |
Displays OCR result. If judgment is enabled, result is shown as OK or NG based on configured rule. |
Time Cost |
Inference time per sample (ms). |
Confidence Threshold |
Minimum confidence for accepted OCR character. Lower-confidence recognition is treated as failure. |
Character Correction |
Constrains first N characters using wildcards: |
Recognition Targets |
Select character types to recognize: uppercase letters, lowercase letters, digits, symbols. |
Concatenation Separator |
Separator between lines when multiple lines are recognized. |
Enable Judgment |
Enables pass/fail verification by character count or content rule (manual pattern or global variable). |
Incremental Training (Optional)
If OCR quality is not satisfactory, use incremental training:
-
Add recognition content: annotate missed or misrecognized characters and provide correct text.
-
Add exclusion content: annotate false-positive background regions.
Then click Train for retraining, return to validation, and verify improvement.
After validation passes, click Save and Use, then Next.