TCGGraph is our first pick for a scan-to-collection workflow that returns a matched printing and prices. Scrydex is another option to evaluate through its Vision feature. We have not independently benchmarked their accuracy or speed.
Which options should you evaluate?
| Option | What to evaluate | Decision detail |
|---|---|---|
| TCGGraph | Recognition has a separate matched-card allowance | Multi-card frames are billed per matched card. |
| Scrydex | Vision is a credit-based API operation | Compare its documented output against your required fields. |
Provider facts come from the linked documentation. Examples and workflow recommendations are this guide’s analysis.
Measure the outcome that matters to a collector
An image can produce the right card name but the wrong printing. Evaluate exact-printing matches, incorrect matches and unresolved images separately. Do not count an unresolved image as a successful identification just because the API returned a valid response.
Use a fixed test set representing your actual capture conditions: sleeves, foils, glare, cropped corners and several cards together. Include negative examples that should be rejected. Keep the sample unchanged when comparing providers.
Design the correction path first
Show the matched image, set, collector number and finish before saving a valuable card. If the response includes alternatives, let the user choose one. Save the corrected identifier alongside the source so later price refreshes remain attached to the accepted printing.
TCGGraph documents confidence thresholds and unresolved results. The appropriate threshold depends on whether a person confirms each match or an unattended process writes directly to inventory.
Budget by images, matches and downstream calls
One photo, one match and one data query are different units. Multi-card images and unsuccessful attempts can affect cost differently across services. Include the calls made after recognition, such as history retrieval, in the workload estimate.
The comparison page breaks down the published billing approaches. Validate actual usage on a small sample before projecting a monthly total.
Keep the recognition evaluation reproducible
Label each test image with the exact intended printing and use the same images for every candidate. Record capture conditions and separate server processing from upload and network time. A scanner can have a fast API response but still feel slow because the app uploads unnecessarily large frames.
Measure what happens after the result as well. Track how many suggestions need manual correction and whether users can complete the save step. For a Pokémon-specific sample and a detailed correction workflow, continue to the Pokémon scanner API guide.
Connect accepted matches to durable holdings
Save the printing that the user accepted, with your own holding ID and ownership details. Keep confidence and original candidate information only where it helps explain or correct the record. An updated provider catalog should not erase the user’s quantity or purchase history.
When several cards are detected, handle each result independently. One unresolved card should not block the confirmed cards, and a partial match should not make the interface imply that the whole image was correctly identified. Use the collection API guide to design the downstream data model.
Frequently asked questions
What is the difference between card recognition and OCR?
OCR reads visible text. A recognition workflow uses the available image information to connect the card to a catalog record. Text can help settle a printing, but a card-name transcription alone is not a complete inventory match.
How should I compare recognition accuracy?
Use a fixed labeled sample and count correct printings, wrong printings and unresolved images separately. Include the capture conditions your users face and report the sample and method with any result.
Are multi-card images always billed as one scan?
No. Providers can meter images, requests or matched cards. TCGGraph documents per-matched-card billing, so one frame containing several matches can consume several scan units.
Sources and further reading
Your next step with TCGGraph
Inspect the documented fields, then test the cards and workload your application needs.