Matrix Filter ?

I think what you are trying to do is a convolution by a kernel matrix (a bit of vocabulary to google here).
To apply it on rgb images, just perform the same computaion on the three componenents channels. As if you where doing what you wrote on three distinct grayscale images.

The kernel you show as an example is the identity kernel, so it will just do nothing.

a better example would be:

|0.09 0.11 0.09||0.11 0.20 0.11| -> some kind of blur|0.09 0.11 0.09|

note that the sum of the elements of the kernel is 1.0 to avoid energy loss (or gain).

I hope this helps,

Phil

hey ^^
ok lets say we have Image array Img[3*3][3] and kernel matrix Km[3*3]
Img [3*3]{3]:
100,250,50 | 125,255,10 | 25,100,88
0 ,100,12 | 20,54,120 | 100,100,100
255,255,255| 50,10,55 | 1,255,0
Km[3*3]:
0 | 0 | 0
0 | 1 | 0
0 | 0 | 0
ok now how can i apply kernel matrix on the bold pixel ?
cya ,

Quote:

Original post by ff8 hey ^^ ok lets say we have Image array Img[3*3][3] and kernel matrix Km[3*3] Img [3*3]{3]: 100,250,50 | 125,255,10 | 25,100,88 0 ,100,12 | 20,54,120 | 100,100,100 255,255,255| 50,10,55 | 1,255,0 Km[3*3]: 0 | 0 | 0 0 | 1 | 0 0 | 0 | 0 ok now how can i apply kernel matrix on the bold pixel ? cya ,

It's just as PixelPhil said, it's applied just like it is to a grayscale image except you're multiplying three channels instead of one. In this particular case it would come out to

(0*...) + (0*...) + ... + (1 * (20, 54, 120)) + ...   = 1 * (20, 54, 120)   = (1*20, 1*54, 1*120)   = (20, 54, 120)

Just post if you need anymore clarification.

thank you both ^^ i have last question ...
let's say the first pixel of the img[0] how can i apply the kernel matrix .. ?
cya ,

You can assume a fixed border color but then ou would have some of this color appearing in you final image.

The cleanest way is to clamp your lookup coordinates between 0 and size-1. this as the effect of propagating the color of the border pixel outside of the matrix.

basically if you need image[-1, 2] take image[0, 2]

example:

      |1 2 3|                |0.3 0   0.1|image |4 5 6|  filter kernel |0   0.3 0  |      |7 8 9|                |0   0   0.3|

image[0,0] = 1 here

the convolution of that pixel is (ignoring the 0s):

0.3 * image[-1,-1] + 0.1 * image[2,-1] + 0.3 * image[0,0] + 0.3 * image[1,1]

because your image is not defined in aras with negative coordinates you transform this to:

0.3 * image[0,0] + 0.1 * image[2,0] + 0.3 * image[0,0] + 0.3 * image[1,1]
= 0.3 * 1 + 0.1 * 3 + 0.3 * 1 + 0.3 * 5
= 2.4

if you hade image[x, 3] you would take image[x, 2] too...

Phil

ah i got it ^_^ thank you vm ^^
cya,

Here is a good page that I found as a good resource.

thank you UltimaX for the link , ^^
i have my last question
ok when i apply kernel matric on the RGB image array let's say this :
.5*(100,255,0) + .1*(255,25,50) + ....=
(.5*100,.5*255,.5*0)+(.1*255,.1*25,.1*50)+...=
the result will be
[(R1+R2+R3+....),(G1+G2+G3+...),(B1+B2+B3+....)] is this correct ??

cya ,