how to slid filter window of 3*3 over the pixels of image.

can anyone plz tell me how to slid a filter window of 3*3 over the pixels of an image one by one.Suppose i have and image with pixels

A= 44 25 22 55 21 11 10 23 ; 35 43 2 23 45 66 86 97 ; 12 22 19 12 76 58 56 23 ; 21 23 43 12 65 75 55 75 ;

I want to slide a filter mask/window of 3*3 over these pixels of image .let the mask be

Mask= 1 1 1 ; 1 1 1 ; 1 1 1 ; plz tell me the coding how to slid this mask over pixels one by one. I have also attached PDF version if anyone know how to do the coading then plz help.

Answer 1

[Mohammad Abouali]

use imfilter() it gives you more option than conv2.

For the border pixels instead of using zeros (as in conv2 case) you can define symmetric, replicate, or even circular boundary condition which gives you better results.

As image Analyst said, Another reason that imfilter is suggested in your case is that conv flips the mask. in your case it doesn't matter because you are using symmetric kernel. if you don't want to flip your kernel you should use cross correlation.

Check here and the graph to see how these two, i.e. convolution and cross-correlation, will produce different output.

http://en.wikipedia.org/wiki/Cross-correlation

Answer 2

[Geoff Hayes]

Ravneet - try using conv2. Using your example from above, we would do

 A = [ 44    25    22    55    21    11    10    23
       35    43     2    23    45    66    86    97
       12    22    19    12    76    58    56    23
       21    23    43    12    65    75    55    75];

 Mask = [ 1     1     1
          1     1     1
          1     1     1];

 C = conv2(A,Mask,'same');

where C would be

 C =

   147   171   170   168   221   239   293   216
   181   224   223   275   367   429   430   295
   156   220   199   297   432   582   591   392
    78   140   131   227   298   385   342   209

The same shape parameter ensures that the output matrix is the same size as A. Note that if the window falls outside of the matrix A, then zeros are used in place (so A is zero-padded).

[Image Analyst]

imfilter() in the Image Processing Toolbox does the same thing and gives you the option of flipping the kernel or not. For this case (all 1's) it doesn't matter but if you ever change the mask Ravneet to something that's not symmetrical, and wonder why the numbers don't match up, then brush up on the definition of convolution, or flip your mask or use imfilter.