Finger vein (FV) recognition is considered a secure biometric trait compared to other biometric modalities, such as fingerprint, face, etc., because it is less affected by external factors such as physiological changes with age and environmental conditions like temperature and humidity. However, with the advancement of deep learning, some spoofing techniques have been developed using advanced generative models, which are an emerging threat to finger vein recognition. Therefore, to improve the reliability of FV falsification detection systems. We have proposed a model that utilizes a Convolutional Auto-Encoder (CAE) to produce realistic counterfeit finger vein images. Our proposed method uses a three-stage approach: (i) a CAE that generates counterfeit FV images; (ii) a second Auto-Encoder (AE) used as a denoising or enhancement model that works as a post-processor to improve the quality of generated images; (iii) and, for the spoof detection task, we utilize multiple machine learning classifiers, namely, CatBoost, LightGBM, Random Forest, and XGBoost, to assess its effectiveness in identifying presentation attacks. The Structural Similarity Index (SSIM) is used to evaluate the quality of generated images, and statistical analysis shows that the image quality obtained with the proposed models (CAE+AE) was significantly better (p-value < 0.05 with a high Cohen effect size) than that obtained with the CAE model alone for image generation. The CatBoost classifier showed superior performance, achieving remarkably low Attack Presentation Classification Error Rate (APCER), Bona-fide Classification Error Rate (BPCER), and Average Classification Error Rate (ACER) values of 0.8%. These results demonstrate that our proposed method improves the security of the biometric finger vein system.