Improving Fairness using Vision-Language Driven Image Augmentation