What Electronic Image Noise Signatures Can Tell Us About Image Linearity and Camera Encoding

Electronic image noise is often described as an artifact of image capture, which typically requires correction. Several sources of electronic image noise contribute to the overall total noise in an image. Some of the noise, known as systematic noise, can be corrected for in-camera, but noise due to specific sources, like photon noise, persists after capture and increases with exposure. We study this relationship between exposure and noise and discover that this relationship reveals information about the linearity, or nonlinearity, of the image encoding. Nonlinear encoding of images through an optoelectronic transfer function (OETF) is a typical image processing step that allows for perceptually efficient use of encoding bit depth, and the accurate measurement or estimation of the OETF becomes necessary for proper processing of images through motion picture post-production workflows. We propose a method based on this relationship and demonstrate how it reveals information that can be used in an inference framework to estimate the OETF and, in turn, linearize images for accurate processing without the need for charts or specific image capture conditions. In addition, we describe the limitations of our method, for example, where extensive and adaptive noise correction can prevent our method from accurately estimating the OETF.