Let's consider the operations which are carried out in spatial area above readout of the digital image (pixels) with the purpose of improvement of quality of this image. Namely the class of operations concerning to a method nonlinear median filtration.
This method is most effective, if noise on the image has pulse character and represents the limited set of peak values on a zero background. The median filter is realized as procedure of local processing by a sliding window of the various form (mask) which includes odd number of readout of the image, and consists that for each position of a window the readout which have got in it are ordered on increase (decrease) of values. Median of examined group from N readout is the average report in this list. This median replaces the central readout in a window for the processed signal. As a result of using the median filter inclined sites and sharp differences of values of brightness of images do not change. This very useful property for images on which, as is known, contours carry the basic information. At the same time pulse noise, which extent makes less than half of window, will be suppressed.
The median filtration has also the lacks. In particular, it is experimentally established, that at that method rather weak efficiency at a filtration so-called fluctuational noise. Besides there is a degradation of contours of the image at increase in the size of a mask and, as consequence, reduction in image sharpness.
The specified lacks of a method can be reduced up to a minimum if take advantage median filtration with the dynamic size of a mask (adaptive median filtration).
The principle of calculation of the central readout at local processing the image by a sliding window remains all the same. This median from a set of the ordered readout which have got in a window (mask), and the size of a sliding window (mask) dynamic also depends by brightness of the next pixels.
Let's enter threshold factor of a deviation of brightness Sthreshold = [0, 1]. Sizes of a deviation of brightness of next pixels A(r, n, m), got in a window in the size n x m, concerning brightness of central readout A (r) will be written down as (1)
Then the criterion according to which it is necessary to increase the size of a mask with the central readout r, will look like (2)
On the basis of the described algorithm the computer program which has confirmed in practice of advantage adaptive median filtrations has been developed.
03/06/2006.
Last projects: Contact lens, Ekaterinburg.
