A Simple Self-Calibration Method To Infer A Non-Parametric Model Of The Imaging System Noise

Imaging system (camera) noise represents a well known source of disturbing artifacts in many algorithms dealing with shape from shading and motion detection, for instance. Some approaches are known to infer the camera noise characteristics. However, all of them rely on a priori assumptions which yield methods depending on the imaging device. In this paper we present a simple method to infer a non-parametric statistical model of the temporal camera noise, based on a inblack-boxli modelling of the imaging system. The model is extracted directly from the pixel intensity variations measured along a short sequence of an arbitrary scene. Extensive experiments accomplished on different sequences acquired with the same camera show that the extracted noise model is strongly scene-independent, thus validating the approach.