Structural health monitoring and biometric security are critical domains benefiting from AI andcomputer vision. This thesis tackles two challenges: generating synthetic data to improve crack detection and enhancing dorsal hand vein recognition through knuckle alignment. For structural crack detection, data scarcity limits AI model effectiveness. Using StyleGAN3 and Brownian Bridge Diffusion Models (BBDM), we generate synthetic crack images with various blur effects to simulate real-world conditions. Integrating this data with hyperparameter-tuned DeepLabV3 achieves 65.62% MeanIoU on the Bridge Crack Library dataset, setting a new benchmark. In dorsal hand vein biometrics, a novel knuckle alignment method enhances vein recognition accuracy, achieving 99.07% on the Jilin University dataset and 99.90% on the Wilches dataset. This study also assesses system robustness under image degradation. Results show that alignment significantly improves performance under moderate blurring but loses effectiveness with severe degradation.