An ROI Extraction Method of Finger Vein Images Based on Large Receptive Field Gradient Operator for Accurate...

Area of Science:

• Biometrics
• Image Processing
• Computer Vision

Background:

• Region of Interest (ROI) extraction is crucial for finger vein recognition preprocessing.
• Current methods using the entire finger region as ROI are limited by background noise and ignore finger position/shape, hindering recognition accuracy.
• Existing techniques struggle with accurate ROI localization in images with large gradient areas.

Purpose of the Study:

• To develop an improved ROI extraction method for finger vein recognition.
• To enhance recognition accuracy by focusing on a specific region and mitigating external image factors.
• To address the limitations of current ROI extraction techniques in handling finger position, shape, and image artifacts.

Main Methods:

• Proposed a novel ROI extraction method by defining a fixed region between two finger joint cavities.
• Introduced a new large receptive field gradient operator inspired by human visual processing for robust target searching.
• Implemented a method to average interference factors over a larger range using a larger size operator to combat noise and uneven illumination.

Main Results:

• The proposed method achieved a significantly reduced Equal Error Rate (EER), with the lowest recorded EER reaching 0.96% across three public datasets.
• Demonstrated effective elimination of influences from finger position and shape on recognition performance.
• Successfully improved the accuracy of subsequent matching recognition by accurately locating finger joint cavities.

Conclusions:

• The novel ROI extraction method effectively enhances finger vein recognition accuracy.
• The technique robustly handles variations in finger position, shape, and image quality.
• The large receptive field gradient operator and averaging strategy contribute to superior biometric identification performance.