Noise-parameter modeling and estimation for x-ray fluoroscopy

In fluoroscopy quantum noise is dominant with respect to other common noise sources, whose effects can be often neglected. Quantum noise is originated by the limited number of photons (low-dose X-ray) involved in fluoroscopic image formation; this noise is commonly modeled as Poisson distributed. Estimation of noise-parameters is required for evaluation of X-ray digital imaging sensors and in several image processing applications (e.g. denoising). The first aim of this work is to validate the analytically derived noise-parameter models by real fluoroscopic image sequences, also considering non-linear gray level mappings currently employed by fluoroscopic devices. A plain procedure for estimation of noise pixel-intensity variance as a function of mean pixel-intensity, which does not require specific test-objects but only some images screening a still scene, has been provided. Besides, a procedure for noise-parameter estimation by differencing fluoroscopic static images has been proposed. It computes image-pair differences, estimates concise parameters of the resulting Skellam distribution and, then, quotes Poisson noise-parameters. Image sequences of a simple step-phantom, acquired with a conventional fluoroscopic device, were utilized for performing the noise measurements. Experimental results confirmed a great agreement of the measured noise-parameters with those analytically derived and the possibility to use static images to estimate noise.