Comparison of Image Denoising using Traditional Filter and Deep Learning Methods
Asavaron Limshuebchuey, Rakkrit Duangsoithong, Mongkol Saejia
Abstract
Image technology often uses in our everyday life. It is useful for many applications such as telecommunication system, automation system in self-driving vehicles, surveillance system and medical research area. However, the image can be interrupted by noise from the environment or electric signal that distorts the detail of the image. In recent years, there had been many image denoising algorithms. Generally, they can be divided into 2 types: traditional filter and deep learning method. This research presents a comparison of the image denoising algorithm using peak signal-to-noise ratio (PSNR) between traditional and deep learning methods on Gaussian noise and Salt and pepper noise condition. Moreover, this experiment also compared the PSNR value of deep learning between noise image to noise image (N2N) learning scheme and noise image to clean image (N2C) learning scheme. According to the results, deep learning method has PSNR value higher than traditional method and N2C learning scheme has PSNR value higher than N2N learning scheme.