Comparison of the impacts of dermoscopy image augmentation methods on skin cancer classification and a new augmentation method with wavelet packets

This work aims to determine the most suitable technique for dermoscopy image augmentation to improve the performance of lesion classifications. Also, a new augmentation technique based on wavelet packet transformations has been developed. The contribution of this work is five-fold. First, a comprehensive review of the methods used for dermoscopy image augmentation has been presented. Second, a new augmentation method has been developed. Third, the augmentation methods have been implemented with the same images for meaningful comparisons. Fourth, three network architectures have been implemented to see the effects of the augmented images obtained from each augmentation method on classifications. Fifth, the results of the same classifier trained separately using expanded data sets have been compared with five different metrics. The proposed augmentation method increases the classification accuracy by at least 4.77% compared with the accuracy values obtained from the same classifier with other augmentation methods.