A test of camera noise models

Models of the sensor noise added in the picture capture process are important components of sensor models for vision systems. Most studies assume that the noise can be described by an additive, independent, identically distributed Gaussian model. This paper presents a test of this hypothesis. It gives the results of applying the test to two camera and A/D converter setups. For both of these, the additive, independent, identically distributed Gaussian model was found to be unacceptable. Results are also presented from an analysis of the experimental data aimed at determining why this is so. After giving the details of the test, the results from applying the test to two systems are presented. Because the hypothesized model was so soundly rejected by the test in both cases, this section contains further discussion to highlight why the model is not acceptable. The paper is concluded with a summary.