Improved Algorithms Research Of Nonlinear Filter In Digital Image

This paper simply introduces the emergence, development, basic working theory and characteristics of nonlinear filter—median filter. Based on these, some improved algorithms of nonlinear filter have been analyzed, according to these, some new improved algorithms of nonlinear filter have been presented in digital image: In the first place, improved multilevel nonlinear weighted mean median filter algorithm has been presented based on weighted and multilevel median filter via joined the nearest four pixels of filter dot in the four filter windows subclass, strengthening the pixels of filter dot as well as its surrounding, stand out the effect of calculating median in the pixels of filter dot as well as its surrounding, making export median close to originality pixel dot even more; Secondly ,an image median filter based on Rough Sets theory has been presented according to several threshold value algorithms of median filter and Rough Sets theory. Via comparing to the contrast between enactment threshold value and the difference of mean value of the filter window and filter dot pixel, applying to the compared result of Rough Sets theory, the image has classified the two kinds of the noise and signal, then noises have only been filtered and reduced the blur of image and held the image detail; Finally, according to the trait that dilation, erosion, opening and closing etc. operation of mathematics morphology can protect image edge, mathematics morphology improved algorithm of nonlinear filter has presented. This algorithm makes the opening operation to the noised image firstly, then closing operation, and does median filter to the difference of the two images operated, after filtered, adds the image to the image of opening operation. This algorithm reduces the image signal loss of the closing and opening operation and keeps the image detail. It has been proved that this algorithm is better than these former algorithms through the contrast of image impersonality criterion (peak signal to noise ratio, normalized mean square error, normalized mean absolute error).