How We Detect
Counterfeit Photocards
BiasTrade uses advanced computer vision algorithms to analyze image textures and detect telltale signs of reproduced or counterfeit photocards.
The Challenge
Counterfeit K-pop photocards are often printed using commercial inkjet or laser printers. These printers use halftone patterns (tiny dots) to reproduce colors, creating a distinct texture that differs from authentic offset-printed cards.
Authentic
Smooth, continuous tones
Counterfeit
Visible halftone dot patterns
Front Side Analysis
Our algorithm analyzes the front of the photocard using a multi-step process designed to distinguish print patterns from normal camera noise.
Center Crop (50%)
We extract the center 50% of the image to focus on the actual card content, ignoring edges and background that might introduce noise.
Gaussian Blur Filter
A 3x3 Gaussian blur removes high-frequency camera noise (like ISO grain from low-light photos) while preserving mid-frequency structural patterns like halftone dots.
Laplacian Edge Detection
The Laplacian operator detects rapid intensity changes (edges and patterns) in the image. Halftone dots create many small edges, resulting in high variance.
Variance Calculation
We calculate the statistical variance of the Laplacian output. Higher variance indicates more texture patterns - a potential sign of counterfeit printing.
Back Side Analysis
The back of photocards typically has large uniform background areas. We isolate these areas to detect CMYK halftone patterns or printer noise.
Otsu Thresholding
We use Otsu's method to automatically separate the background from text and logos. This creates a mask that isolates only the background pixels for analysis.
Authentic cards have very smooth backgrounds. Counterfeits often show texture from CMYK printing or inkjet/laser printer artifacts.
Compression-Resistant Analysis
The Problem: When images are shared on social media or messaging apps, they get heavily compressed. This removes the fine halftone patterns that texture analysis relies on. Counterfeiters can exploit this by sharing compressed images.
Our Macro Analysis detects printer artifacts that survive compression by analyzing large-scale color and tonal characteristics instead of fine texture details.
Gradient Banding Detection
Authentic photocards have smooth, continuous color gradients. Printed fakes often show visible "stepping" or banding in gradients due to limited printer color depth. We analyze color transitions to detect these stair-step patterns.
Black Point Depth (L* Analysis)
Authentic offset printing achieves deep, rich blacks. Consumer printers produce "lifted" blacks that appear grayish. We measure the darkest pixels in the image using L* (lightness) from the Lab color space.
Color Cast Detection
Inkjet and laser printers often produce subtle color casts - a shift toward cyan, magenta, or yellow in areas that should be neutral. We analyze supposedly neutral tones (grays, whites) for unwanted color bias.
Macro Analysis Scoring (0-100)
No suspicious macro-level patterns detected.
Some patterns detected. Consider other indicators.
Strong indicators of printer output detected.
Why it works: These macro-level characteristics are preserved even after heavy JPEG compression because they affect overall color distribution rather than fine pixel-level details.
Ownership Verification
Beyond detecting counterfeits, we verify that sellers actually own the photocards they list. This prevents scammers from downloading images from the internet and pretending to sell cards they don't have.
How It Works
Sellers upload a verification photo showing the photocard alongside a handwritten note with their username and current date. Our AI then compares this to the original listing photo.
2024-01-15
Same card detected. Ownership verified.
Different cards. Potential fraud alert.
Neural Network Embeddings
We use MobileNet to extract visual features from both images, creating compact "fingerprints" (embeddings) that capture the essence of each photocard. Cosine similarity between embeddings determines if they show the same card.
Three-Zone Scoring
Our scoring system categorizes results into three zones for easy interpretation.
Front Side Thresholds
Normal image texture. No signs of print patterns.
High noise detected. May be due to low lighting. Retake recommended.
Halftone patterns detected. Strong indication of counterfeit.
Back Side Thresholds
Smooth background. Consistent with authentic cards.
Some texture in background. Could be lighting or quality issue.
CMYK halftone or printer noise detected in background.
Technology Stack
OpenCV.js
Computer Vision
WebAssembly
Native Performance
Client-Side
Privacy First
Real-Time
Instant Results
Important Notice
This technology provides an estimate based on image analysis. Results may be affected by photo quality, lighting conditions, or camera settings. For high-value transactions, we recommend using multiple verification methods including physical inspection.
Test our fake detection technology with your own photos