Measuring the Noise of Digital Imaging Sensors by Stacking Raw Images Affected by Vibrations and Illumination Flickering

International audienceThis paper discusses camera noise estimation from a series of raw images of an arbitrary natural static scene, acquired with the same camera settings. Although it seems natural to characterize noise from the random time fluctuation of pixel intensity, it turns out that these fluctuations may also be caused by illumination flickering and mechanical microvibrations affecting the camera. In this context, the contributions are twofold. First, a theoretical model of image formation in the presence of illumination flickering and of vibrations is discussed. This parametric model is based on a Cox process. It is shown that illumination flickering changes the standard affine relation between noise variance and average intensity to a quadratic relation. Second, under these conditions an algorithm is proposed to estimate the main parameters governing sensor noise, namely the gain, the offset, and the readout noise. The rolling shutter effect, which potentially affects the output of any focal-plane shutter camera, is also considered. Experiments show that this simple method gives results consistent with the photon transfer method, which needs a special experimental setting and several data acquisitions, and with an algorithm based on a single image. The main practical result is to show that flickering, which is generally considered as an artifact, here plays a positive role since it finally enables us to estimate any of the sensor parameters