Burst image denoising

Abstract

Modern smartphone images go through very heavy image processing before they are, for example stored, transmitted or presented on the screen. Image denoising, a process of removing noise, is one of the very first steps in smartphone image processing. Image denoising is an important step in smartphones because unprocessed images have relatively high noise levels, which can make images visually less appealing. The high noise levels are mostly caused by the relatively small sizes of the smartphone cameras and the restrictions on the exposure time. High exposure time decreases the noise levels, but also increases the blurriness of the image due to motion blur, and therefore low exposure times are often preferred.

The performance of the image denoising methods have only seen incremental improvement over the years. Also, in recent years most of the flagship smartphones have begun to use burst denoising methods. Burst denoising methods combine several images into one such that resulting image has greatly decreased noise level.

In this thesis we evaluate how burst image denoising methods compare to more traditional single-image denoising methods and what kind of future burst denoising methods hold. We also explain what the image capture process in smartphones consists of, and how both single-image and burst image denoising methods operate.

We also do experimental testing to show that burst denoising methods have a high potential in denoising performance and that they can result very good denoising results in smartphones.