Estimation of quantum noise in fluoroscopy by analyzing differences of static images

Quantum noise is typical of images obtained with a small amount of photons and X-ray fluoroscopy is an example of the type. Indeed, to keep the doses of radiation to acceptable levels for the patient and, at the same time, to allow prolonged screening, the number of X photons is drastically reduced giving rise to a remarkable quantum noise on the images. For various applications it is necessary to estimate the level of this noise: this study proposes to use a limited number of fluoroscopic images depicting a static scene to estimate the noise. By considering the noise samples uncorrelated (e.g. for a single pixel) and computing all possible differences due to the number of images, it is possible to estimate fairly accurately the characteristics of the noise (and the gain of the sensor). Statistically, quantum noise is Poisson distributed, while the difference of two Poissonians gives a Skellam distribution. Making a difference between two static images will automatically result in cancellation of the image (the scene), whatever it is, highlighting only the noise. Using only few images it is possible to provide an estimation of noise, which is comparable (only slightly less accurate) to that should be obtained from an analysis performed with much more images.