Finger Vein Recognition by Combining Anisotropic Diffusion and a New Feature Extraction Method

Abstract

In recent years, Finger Vein (FV) Recognition System is frequently used where personal security is required. Image distortion caused by light scattering in the tissue is one of the major problems about the visibility of the FV. In this study, Homomorphic Filter and Anisotropic Diffusion are used for removing the light scattering problem in our captured FV image and to increase the visibility of the veined region. Novelty of the study is proposing two new features: Horizontal Total Proportion (HTP) and Vertical Total Proportion (VTP). These two new features were used together with both spatial and frequency domain features and it was observed that the success rates obtained by our attributes were significantly increased. Experimental results demonstrate that the proposed HTP and VTP features are effective and reliable to improve the classification success in FV recognition problem. According to the experiments, the use of Perona-Malik and Homomorphic Filter together has been shown to reduce the light scattering problem and improve vascular visibility by removing the noise in the finger vein image. In this study, four different classifiers are used: Complex Tree, Ensemble, Support Vector Machines (SVM), K-Nearest Neighbors (KNN). The best success rate was achieved by using the KNN classifier.