OCR labels: camera and character sampling

Purpose and operating boundary

Capture readable source characters before configuring OCR, while preserving stroke shape, field margins and acquisition identity at the required movement conditions.

This tutorial defines an engineering evaluation, not a ready-to-run product certification. The current Deyi Vision public PC example covers an EAN-13 workflow on confirmed sample images. Native OCR and detection work in the engineering code must be checked against the installed runtime, bound models and job before use. A browser template does not prove browser execution: the planned OpenCV.js backend needs separate verification. Cloud recognition requires explicit approval of the image scope, provider, model and cost. A review archive is not an executable release package. Camera, firmware and driver compatibility require their own check. All worksheet values below are hypothetical; they are not Deyi Vision measurements or customer results.

Prepare the evidence

Use labels with actual font, print process and label curvature. Include confusing characters, faint strokes and the smallest declared text field. Before reusing the capture configuration, check that the stored pixel format and active region still match the saved recipe. Record any binning, resizing or image conversion because the object scale calculated from source pixels no longer applies unchanged after those operations. Compare the archived frame with the frame passed into the first processing node. If they differ, retain both identities and explain the conversion. During a reconnect trial, distinguish a new exposure from a repeated buffered frame. Disable automatic pass-through while acquisition identity is uncertain, and retain the event that caused the mismatch rather than silently resetting the counter. OCR label verification is a data-integrity task rather than a search for plausible words. Define each field, its location, expected alphabet, exact comparison policy and the trusted source of the expected value. A lot field, date field and product name may need different rules. Keep the text detection polygon and recognizer output before normalizing spaces or case. Do not silently replace I with 1 or O with 0: that can make a wrong label appear correct. The original engineering evidence includes a narrow OCR boundary case that returns NoResult or Review; increasing a crop does not establish general recognition readiness. A readable field can still contain the wrong lot, while a correct expected value does not justify filling an unreadable field with that value. Begin with an acquisition record rather than a displayed preview. A preview may be resized, sharpened or reused after the input has stopped. Save the unmodified frame, image dimensions, pixel format, exposure, gain, lens setting, working distance, trigger identity and acquisition time. Include the smallest relevant feature and the largest allowed position variation in the same test. First capture a stationary sample to separate focus from motion blur; then repeat at the actual movement conditions without changing the optical geometry. Link every saved image to the physical sample or event. A missing or stale frame is an acquisition failure even when the last visible image looks good. Start a Deyi Vision area-scan evaluation with monochrome capture when characters and substrate separate reliably in intensity. Compare a color route when print and background differ in color but have similar grayscale values; verify the actual channel or grayscale conversion supplied to OCR. Frame the complete text field and smallest strokes at the allowed label curvature. Fix focus, exposure, gain and illumination, then test moving labels with the actual shutter and trigger configuration. Confirm camera interface, driver and installed OCR runtime independently; selecting a camera does not establish recognizer readiness.

Procedure

1. Measure source character height and stroke width at the object plane, not after preview enlargement.
2. Focus the label across the permitted height range and compare stationary and moving captures with the same lighting.
3. Limit exposure from line speed and object sampling; preserve character gaps instead of compensating blur with sharpening.
4. Store the original field with surrounding margins, expected lot and image ID; do not resize an unreadable sample into claimed recognition evidence.

Interpret the trial

At the hypothetical scale, movement during exposure is 0.5 px. Character height of 20 px is a trial observation, not a guaranteed OCR minimum. A short field can still be unobservable even if the overall image dimensions are accepted.

Recording worksheet

s = 40 / 1000 = 0.04 mm/px
blur = 200 * 0.0001 / 0.04 = 0.5 px

Two engineering views

Failure diagnosis

• Joined strokes only during movement indicate exposure or timing; compare the static original.
• Missing character tops indicate cropping or pose; expanding a preview does not restore lost source pixels.

Acceptance and handover

Accept the acquisition when the declared fields and difficult characters remain observable across print and position variation. Hand over both readable and unreadable originals; OCR accuracy is evaluated later.

Keep the source images, the expected decision, the exact revision and the complete result record together. Freeze the evaluation set before selecting the final parameters. If those answers are later used to tune the workflow, mark that set as development data and obtain a new independent acceptance set. Keep Review, NoResult, execution error and missing acquisition visible in the denominator; do not turn them into passes by filtering rows. Save a reproducible failure alongside the successful sample. The handover should identify who may change the recipe, how a prior approved revision is restored and which device output remains inhibited until acceptance. A software result alone does not establish that a moving physical part reached the correct destination.