Leveraging Unsupervised Learning for Cost-Effective Visual Anomaly Detection