Dorsal Hand Vein Pattern Recognition: A Comparison between Manual and Automatic Segmentation Methods

Area of Science:

• Biometrics
• Computer Vision
• Medical Imaging

Background:

• Dorsal hand vein (DHV) pattern extraction is crucial for biometrics.
• Existing segmentation techniques often suffer from small datasets and inconsistent results.
• Manual segmentation is time-consuming and prone to variability.

Purpose of the Study:

• To compare manual segmentation with a novel hybrid automatic segmentation method (HHM) for DHV pattern extraction.
• To evaluate the performance of deep learning models trained with different segmentation strategies.
• To assess the feasibility and efficiency of the proposed HHM for DHV image analysis.

Main Methods:

• A hybrid automatic segmentation method (HHM) combining histogram equalization, morphological operations, and thresholding was developed.
• The Bosphorus dataset was used, with manual segmentation serving as ground truth.
• AlexNet was trained on segmented and augmented dorsal hand vein images, with data split into 8:1:1 for training, validation, and testing.

Main Results:

• A model trained with manual segmentation achieved 91.5% test accuracy.
• The HHM method alone yielded 76.5% test accuracy.
• Including automatically segmented and augmented images improved test accuracy to 84%, with low false acceptance (0.00035%) and false rejection (0.095%) rates.

Conclusions:

• The proposed hybrid automatic segmentation method (HHM) is feasible for dorsal hand vein (DHV) region-of-interest extraction.
• This automated strategy offers improved consistency and efficiency compared to manual segmentation.
• The technique demonstrates competitiveness with existing methods in DHV pattern extraction.