무작위 주파수 필터링을 이용한 데이터 증강 기법

초록

In this paper, we propose an algorithm for augmenting data in a frequency domain applying Fourier transform to an image. The algorithm proposed in this paper proposes to convert an image into a frequency domain and then convert the image with random masking into the frequency domain and use it as new data. The data used for this paper are ISIC 2018 skin cancer object segmentation dataset[1]. Based on this data and the data augmentation algorithm proposed in this paper, the segmentation performance was measured based on the Dice Coefficient. The original data set was 0.79, and the algorithm proposed in this paper achieved a performance of 0.87.