Today, finger vein identification is gaining popularity as a potential biometric identification framework solution. Machine learning-based unsupervised, supervised, and deep learning algorithms have had a significant influence on finger vein detection and recognition at the moment. Deep learning, on the other hand, necessitates a large number of training datasets that must be manually produced and labeled. In this research, we offer a completely automated unsupervised learning strategy for training dataset creation. Our method is intended to extract and build a decent binary mask training dataset completely automated. In this technique, two optimization steps are devised and employed. The initial stage of optimization is to create a completely automated unsupervised image clustering based on finger vein image localization. Worldwide finger vein pattern orientation estimation is employed in the second optimization to optimize the retrieved finger vein lines. Finally, the proposed system achieves 99.6 - percent pattern extraction accuracy, which is significantly higher than other common unsupervised learning methods like k-means and Fuzzy C-Means (FCM).