fuzzy logic for image processing

Dear all,

       I want to use fuzzy logic for image processing. But I've following problem.
1) How to transfer the input image for fuzzy logic functions.
I need to evaluate each pixels in a particular area.

Thank you.

[nayomi ranamuka]

Furthermore I need r,g,b components separately for determining membership function. Can anybody help me ?

[jagkaran]

u can gather your info from http://www.icgst.com/gvip/volume9/issue5/p1150913709.pdf

RGB components can be seperated by using ...

red = image(:, :, 1);

green = image(:, :, 2);

blue = image(:, :, 3);

image can be noisy or original ....

Answer 1

[Vito]

"Fuzzy logic" is one of variants of neural networks.

 From - for restrictions, for image processing doesn't approach.

"Fuzzy logic" it is equivalent to statistics.

"Membership Functions"-functions of distribution probability.

Therefore the result which is received on an output, no more than a probability estimation of input values.

Considering that result of any logical expression - one value (for sample: (a) or not (a) =true), application "Fuzzy logic" as an image handling facility is extremely restricted.

If the task - an estimation pixel in area "Fuzzy logic" can be used.

Let's admit, the image represents RGB.

The array from three values - R, G, B will be input.

input (R, G, B)-> output (estimation).

But on an output, there will be a probability allocation of these values to all set.

If it is necessary to estimate in some area it is necessary to break set into classes (for example: a red class, a green class, a blue class).

The first network (anfis1).

               R G B output

trnData = [255 0 100 (R*1+G*2+B*3)/1530=355/1530=0.23;

             50 255 0 (R*1+G*2+B*3)/1530=305/1530=0.1993;

             0 200 255 (R*1+G*2+B*3)/1530=455/1530 = 0.2974;];

........ We receive fuzzy sets (areas).

Output defines area and pixel. The estimation shows a level of product of sets R, G, B.

Changing, after training value of any of input parameters, we will receive a pixel estimation (and accordingly area).

output (anfis1) = range [0, 1];

That is we can analyze the image about color blending.

For example.

estimation = 1 = (R or G or B) = white color (perfect square)

estimation = 0.5 = (R or G or B) = gray color (half of square)

It is possible to add and the second network.

The second network (anfis2).

trnData = [output (anfis1) R G B estimation = (R*1+G*2+B*3)/1530]

output (anfis2) = range [0, 1];

Changing value output (anfis1), we will receive various points in once square.

------------- |

| A |output (anfis2)

| |

| _ ______ |

output (anfis1)

A (x=output (anfis1), y=output (anfis2)) = A (0.5,0.5) will correspond to pixel having gray color.

But it is better to use powerful neural networks.

[Dil]

Can you explain how can I link the output from image processing to fuzzy controller model? For example say we calculate some distance between two objects in the image. how can i set this distance as an input to the fuzzy controller? thank you...

Answer 2

[Arkadiy Turevskiy]

This new example might help: Fuzzy Logic Image Processing .