Unsupervised learning based noise reduction algorithm in 2D Rayleigh images

Abstract

This work reports a novel image denoising and reconstruction algorithm based on unsupervised learning for removing Mie scattering interference in Rayleigh images. We first superimposed numerically simulated Rayleigh images and noise images acquired in experiment to generate noisy Rayleigh images as training data. The proposed unsupervised model was then trained based on unpaired datasets. Finally, extensive evaluations were conducted to demonstrate a convincing denoising result, which displayed an excellent reconstruction quality with a peak-signal-to-noise of ~41dB and an overall reconstruction error of ~0.5%. The results showed that our algorithm was able to provide an alternative method for noise reduction in two dimensional Rayleigh measurement of combustion.