An expectation maximization approach to impulsive noise removal in digital radiography

Objective Impulsive noise constitutes a serious disturbing factor in digital radiography. The aim of this paper is to propose a new filter which is capable of eliminating efficiently this kind of noise. Method The filter is based on the classical switching scheme: the pulses are first detected and then corrected through a median filter. The novelty element is a pulse detector based on an adequate statistical model of the image noise. This model is constituted of a mixture of photon counting and impulsive noise. Results The filter operation has been verified and compared with existing algorithms on both synthetic and real images. The filter is capable to remove, on the average, more than 94% of the pulses, with a positive predictive value higher than 80%. The computational time required to filter a radiograph of 4.5 Mpixels is just above 2 s. Conclusion Thanks to the accurate description of the noise statistics and the efficient computational scheme implemented, the filter can be used to reliably and quickly remove impulsive noise from digital radiographs. The experimental results demonstrate the superiority of the proposed filter with respect to traditional methods.