Image Denoising with Self Operational and Convolutional Cycle-GANs
Hodaka Yamanouchi, Yusuke Sao, Toshiyuki Uto · 2023
In this paper, we propose self operational and convolutional cycle-consistent generative adversarial networks (SOC-Cycle GANs), which are low learning models for removing noise in medical images. We focus on operational Cycle-GANs that combine Cycle-GANs and self operational neural networks (Self-ONNs), and by incorporating convolutional neural networks (CNNs) into the operational Cycle-GANs to reduce the load on the operational Cycle-GANs model. In addition, we eliminate noise from medical images using the SOC-Cycle GANs. In this research, we train and experiment with the SOC-Cycle GANs on a dataset of X-ray images, where the SOC-Cycle GANs are trained on 500 X-ray images with noise and 500 clean X-ray images without noise. In the simulations, we conduct a quantitative evaluation in terms of PSNR and SSIM value, number of parameters, and training time. We show that the our proposed method is model that do not require much training time and are effective methods for removing noise from medical images.