Data augmentation is important for balancing datasets and increasing dataset sizes, in case of a shortage of samples and/or imbalanced classes. In the present paper, we explore two avenues for image databases enhancement through embedding vector fields (VFs). In the first avenue, we augment the image features with VF features. This technique makes the images more distinct, which improves the skin lesion classification. In the second avenue, we use the images with embedded VF features to augment the original image databases. We experimentally validate the capabilities of the proposed augmentation techniques to improve machine learning (ML) classification. Hence, we use the public skin lesion database ISIC 2020 and conduct the experiments with five neural networks (NNs) to compare classifications on the original ISIC 2020 and its versions with embedded VFs. We show that the two kinds of augmentations (image features and image databases) improve the ML classification of skin lesion images.