MAGE-Vein: Multi-Instance Age and Gender Estimation from Finger Vein Images

Finger vein recognition is a standard biometric technology, but its reliability is often questioned due to template aging—the physiological changes in veins over a lifetime. Previous research concluded that age estimation from finger veins was impractical, largely due to overfitting on demographically biased public datasets. This paper investigates whether accurate age estimation is possible when these biases are removed and when physiological confounding factors, such as gender-related vascular differences, are explicitly addressed.

The authors propose MAGE-Vein, a multi-task learning framework that processes images from three fingers (index, middle, and ring) of the same subject. The model uses a DenseNet-161 backbone to extract features from each finger, which are then integrated using a hybrid fusion strategy—combining concatenation and averaging—to suppress local imaging noise. Crucially, the network is trained to perform both age regression and gender classification simultaneously. This multi-task approach forces the model to learn gender-invariant features, effectively separating structural aging signatures from gender-specific vascular traits like vessel diameter and hemoglobin concentration.

By utilizing a demographically balanced dataset of 402 subjects with precise chronological age labels, the authors demonstrate that the previous consensus on the impracticality of finger vein age estimation was an artifact of poor data. MAGE-Vein achieves a Mean Absolute Error (MAE) of 6.12 years and a correlation of 0.880. The ablation studies confirm that feature-level fusion across multiple fingers significantly outperforms single-finger or score-level fusion, and that multi-task learning with gender classification is essential for achieving high accuracy.

This work challenges the prevailing view that finger vein patterns are static or that age estimation is impossible in this modality. By providing a robust framework and demonstrating that high-quality, balanced data is the primary driver of performance, the authors open new possibilities for using finger veins as a source of soft biometric information, which could enhance the security and adaptability of existing biometric systems.