GB/T 35676-2017 in English

This standard constructs a complete finger vein recognition algorithm evaluation system, which mainly includes three technical modules:

• Test library construction specifications: stipulates a sample size of more than 1,000 people, and must cover demographic characteristics such as gender, age, and region
• Performance indicator system: defines core indicators such as FAR (false acceptance rate), FRR (false rejection rate), and EER (equal error rate)
• Test methodology: details the feature file library generation process and the two-stage testing process

Test library construction specifications

Library construction must strictly follow the following:

1. Sample size ≥ 1000 people, with at least 2 fingers per person (1 on each hand)
2. Collect ≥ 5 images for each finger, in line with the GB/T 35742 image standard
3. Population distribution must be balanced: 1:1 for gender, 1:1 for north and south, and 1:1 for urban and rural population
4. Age stratification: 18% for those under 15, 65% for those aged 16-59, and 17% for those aged 60 and above

Two-stage test flow

Non-embedded system requirements

The dynamic link library needs to implement 5 core functions:

Embedded system protocol

The SCSI-3 specification extended instruction set is adopted. The key commands include:

• 0x03: Feature extraction command (H×W bytes of image data need to be transmitted)
• 0x04: Feature matching command (returns a similarity value of 0-1000)
• Error requirement: cross-platform similarity deviation ≤0.001

This standard reflects the three major development trends of finger vein recognition technology:

1. Multimodal fusion: Ensure the stability of feature extraction through EFR indicator control
2. Real-time optimization: Response time indicators drive algorithm efficiency
3. Cross-platform compatibility: Standardized interface design for embedded and non-embedded systems

Typical application case: A bank adopted the finger vein ATM system tested by this standard and achieved commercial-grade performance of FAR≤0.01% and FRR≤1.5%.

Enterprise Implementation Path

• Hardware Selection: Prioritize embedded devices that support USB3.0 protocol
• Algorithm Optimization: Focus on reducing the EER value, and the recommended target is ≤0.5%
• Test Verification: Must include benchmark tests under 0%EFR conditions

Evaluation Method

When comparing algorithm performance through DET curve:

1. Lower FRR under the same FAR is better
2. The smaller the area under the curve, the better the overall performance
3. Focus on the performance in the 0.1%-1%FAR range