Statistical Calibration of the CCD Imaging Process

Charge-Coupled Device (CCD) canieras are widely used imaging sensors in coniputer vision systems. Many photometric algorithms, such as shape from shading, color constancy, and photometric stereo, iniplicitly assume that the image intensily is proportional to scene radiance. The actual image measurements deviate significantly from this assuniption since the transforniation from scene radiance to image intensity is non-linear and is a function of various factors including: noise sources in the CCD sensor, as well as various transformations occurring in the canieru including: white balancing, gamma correction and automatic gain control. This paper illustrates how careful modelling of the error sources and the various processing steps enable us to accurately estimate the “response fiinction ”, the inverse mapping from iniage measurements to scene rudiunce for a given camera exposure setting. It is shown that the estimation algorithm outpe@ornis the calibration procedures known to us in ternis of reduced bias and variance. Further, we demonstrate how the error modelling helps us to obtuin uncertainty estiniuies of the camera irradiance value. The power of this uncertainty modeling is illustruted hy a vision task involving High Dynaniic Range image generution followed by change detection. Change can be detected reliably even in situation where the two images (the reference scene image and the current image) are taken several hours apart.