1、Improved 2-D Median Filter for On-Line Impulse Noise Suppressiom Abstract-An inproved 2-D median filter employing multishell concept to suppress impulse noise ,is presented.The performance of proposed filter is evaluated over image LENA ,The impulsive noise is added using MATLAB utility.The modified
2、 strategy reduces the mnuber of replacement and results in better performance and simple hardware realization that is suitable for on-line implementation. Index terms-Median Filter , Multi-shell Median Filter, Impulse Noise I.INTRODUCTION In TV and other imaging systems,impulse noise is a common imp
3、airment . The standard T.V.Broadcast signal is often contaminated with impulsive noise arising from various sources such as household electrical appliance and atmospheric disturbances.Broad banding of the signal further increases the level of impulsive noise. Various filters are proposed to suppress
4、 such impairments1.The median filter(MF)1-2 is widely used for impulse noise suppression and the multishell median filter(MMF)3 introduces the concept of missing line recovery. Although these filters have satisfactory performance, MMF failsto filter two impulse noises in the same prossing window. Mo
5、veover,these filters tend to blur the images due to too many replacements. C.J.Juan proposed a modified multishell median filter (MMMF)4, which removes most of the shortcomings associated with the MF and the MMF. However, it is observed that under certain condions, to be discussed in the follow sect
6、ions, MMMF fails to perform the desired filtering operation .Moreover,the number of calculations/replacements invoved on the basis of MIN/MAX conditions is still too large and makes the filter difficult to realize,particulariy for real time applications. In this paper, the threshold strtegy of MMMF
7、is modified so that: (a) effective noise filtering operations are performed under all conditions,and (b) number of calculations/replacements is reduced and simplified. This results in a simple hardware realization of the filter. II.PROPOSED MODIFICATION Consider a 3x3-processing window, with P5 as t
8、he central pixel,as shown in Figure 1. P1 P2 P3 P4 P5 P6 P7 P8 P9 Fig.1. A 3x3 processing window The output of MMMF as proposed in 4 is Output (X,Y)= Max(P2,P8)if P5 MaxS P5 if Min s MaxS Min(P2,P8) if P5 MaxS (1) Where S is the set of samples surrounding central pixels except(P4.P6)i.e. S= P1,P2,P3
9、,P7,P8,P9 (2) The principle invoved in the replacement strategy of Equation (1) is that if P5 is corrupted by noise ,it is better to replace its gray level by P2 or P8 than by using MinS or MaxS .also,due to missing lines error,since P4 and P6 may be lost, they are not considered in Equation(2). The
10、 limitation of Equation(1) is that when MinS or MaxS are also corrupted by impulse noise,i.e.either MinS or MaxS is equal to P5,Equation(1)fails to perform the desired filtering operation.To overcome this limitation following modifications in the replacement strategy of Equation(1),are proposed. Output (X,Y)= Max(P2,P8)if P5 MaxS P5 if Min s P5 MaxS Min(P2,P8) if P5 MaxS (3) It has been observed that more than 70-80% points in an image,the gray level diatances of P5 from(P2 or P8) and from MaxS are below 16.This is shown in Fig.2 for the image
