A new switching median filter for digital radiography

Localized hardware failures on sensor and communication channel often introduce in digital radiographies a characteristic impulsive noise, known as "salt & pepper". Eliminating these failures, which generate the corrupted pixels, is a very costly or even impossible task; therefore, image processing techniques have been developed to correct the gray level values of these pixels. We propose here a new switching median filter for digital radiography, which has been inspired by the properties of the human visual system: a pulse is detected on the radiography by measuring its visibility in terms of local contrast and signal to noise ratio. The local background value is estimated applying a 3×3 median filter to the image, and noise distribution is evaluated by means of a simple and reliable model, which describes the properties of the digital sensor and incorporates the statistical characteristic of the noise. Since the filter requires no iteration, it can work in real time also on large images (less than 0.7s for 12 bit, 4.8 MPixel images). The presence of edges does not affect the pulse detection and correction, thus resulting in a more efficient approach with regards to more traditional methods. Some residual pulse may remain visible only in the darkest zones of the radiography, which are usually poor of diagnostic information.