Finger Vein Segmentation in Infrared Images Based on Model Fitting
In this paper, a novel method for finger vein segmentation in infrared images based on a model that represents the intensity distribution along a cross-sectional profile of vein, is proposed. The method is based on model fitting of brightness intensities over a local neighborhood. A simple second-order degree polynomial is used in order to represent the brightness intensities along the cross-sectional profile of veins. Two different approaches are adopted. According to the first, a multiscale multidirectional model fitting is performed based on the assumption that the second-order derivative on a vein pixel is positive. The second is based on multiscale model fitting of intensities in a single direction using two assumptions about the intensities of the profile of veins. According to the first assumption, the second-order derivative is positive on a vein pixel and according to the second, the first-order derivative is zero on a vein pixel which means that the pixel belongs to vein centerlines. The proposed method is robust, in both approaches, as the experimental results show and it achieves high evaluation rates in terms of sensitivity, specificity, and accuracy.