Adaptive filtering in magnetic resonance images

It is shown that the conventional filters as mean, median and sigma filters can be quite simply improved with the usage of thresholds for standard deviation (sigma) in a window, that is by using an adaptive window size. This improvement helps to overcome the major drawbacks of conventional filters, namely blurring of object boundaries and the suppression of the fine structural details. Filter thresholds are estimated from a noise histogram of an image. The noise histogram is derived from several most homogeneous windows (not from just the one most homogeneous window as usual), the number of which is controlled by the noise thresholds. A new filter quality measure, the noise histogram difference, is introduced which allows one to evaluate the smoothing strength and edge preserving of filters quantitatively