Documentation
AI Card Recognition
SuperCatch uses a 3-step recognition pipeline to identify cards and limit unnecessary manual review.
Page Guide
Jump directly to the sections that matter for this page.
Recognition Order
Database -> OCR -> AI
Cheaper and faster matching paths run before the vision model is used.
Stop Condition
High Confidence
The pipeline stops as soon as confidence is strong enough for review.
Cost Control
Free + Low Cost First
AI Recognition stays targeted instead of running on every upload by default.
Recognition Pipeline
When you upload a card image, our system processes it through multiple stages, stopping as soon as it achieves high confidence.
Database Auto-Matching
FreeOur system first attempts to identify front/back orientation and match against known card images:
Front/back detection from filename patterns ("-front", "-back", "-rear")
Image hash comparison against known card images (coming soon)
Visual feature matching using computer vision (coming soon)
✓ If confidence is above 85%, the process stops here.
OCR Text Extraction
Low CostIf database matching fails, we use Optical Character Recognition (OCR) to extract text:
Extract player names, years, set names, and card numbers
Identify special features (rookie cards, autographs, serial numbers)
Detect grading information (PSA, BGS, CGC, SGC grades)
Match extracted text against our card database
✓ If confidence is above 75%, the process stops here with structured card data.
AI Recognition (Last Resort)
Higher CostIf database and OCR steps do not produce a strong match, we use AI Recognition:
SuperAI vision analyzes the entire card image
Extracts card details that were not resolved earlier in the pipeline
Enhanced with database matching for improved accuracy
Cost monitoring keeps AI usage targeted
✓ This step runs only when earlier recognition steps do not produce a strong match.
Front and Back Card Detection
SuperCatch intelligently groups front and back images of the same card:
Filename analysis: looks for "-front", "-back", or "-rear" indicators
Automatic pairing: first image = front, second image = back (if no filename indicators)
Smart grouping: matches cards with similar attributes but different sides
Visual indicators: shows "Front & Back Detected" with thumbnail previews
Recognition Method Indicators
Each recognized card shows which method was used:
Database auto-matching
OCR Text Extraction
AI vision recognition
Requires manual entry
Why This Approach Works
Cost Efficient
Free and low-cost steps run first so AI usage stays targeted and review stays efficient.
Fast Results
Database matching resolves known cards first, while OCR handles many of the remaining cases before AI Recognition is needed.
High Accuracy
The 3-step pipeline uses the most appropriate recognition step for the card data available.
Section
Recognition Tips
Photo and Upload Tips
📁 Use Descriptive Filenames
Name your files like "2023-topps-chrome-mike-trout-1-front.jpg" for instant database matching
💡 Clear, Well-Lit Photos
Ensure text is readable for optimal OCR performance
🔄 Include Both Sides
Upload front and back images for complete card information
📷 Supported Formats
Use JPEG, PNG, or WebP images up to 10MB each
Review Detected Data
Review every detected field before publishing: you can edit card data, condition details, and any recognition result manually before the listing goes live.
Related Guides
Explore the adjacent workflows that usually sit next to this topic.