I have a multispectral image (256*256*6)and hyperspectral image(64*64*191) how can I divide the hyperspectral image into six groups?

I want to make a fusion of the hyperspectral image and multispectral image, so I want to use the giuded filter one of the steps procedure, how can I divide the hyperspectral image into six groups?

Answer 1

[Walter Roberson]

You have not indicated anything about how the images are to be divided into groups, so we will arbitrary decide that what you need is to divide the images into groups of consecutive panes to minimize the squared difference between the group and the mean of the group.

One way to find the best partitioning for this purpose is to use genetic algorithm:

A = your_hyperspectral_image;

V = @(M) M(:);
P = size(A, 3);
R = @(x) round(x);
obj = @(L) var(V(A(:,:,1:R(L(1)))))*R(L(1)) + var(V(A(:,:,R(L(1))+1:R(L(2)))))*(R(L(2))-R(L(1))) + var(V(A(:,:,R(L(2))+1:R(L(3)))))*(R(L(3))-R(L(2))) + var(V(A(:,:,R(L(3))+1:R(L(4)))))*(R(L(4))-R(L(3))) + var(V(A(:,:,R(L(4))+1:R(L(5)))))*(R(L(5))-R(L(4))) + var(V(A(:,:,R(L(5))+1:end)))*(P - R(L(5)));
ca = [1 -1 0 0 0;0 1 -1 0 0;0 0 1 -1 0;0 0 0 1 -1];  %linear constraints to ensure partition is sorted
cb = [-1 -1 -1 -1];
cAeq = [];
cbeq = [];
lb = [1 2 3 4 5];
ub = P-4:P;
cnonlcon = [];
options = optimoptions(@ga, 'Display', 'iter', 'PopulationSize', 2000);
[part, residue] = ga(obj, 5, cA, cb, cAeq, cbeq, lb, ub, cnonlcon, [1 2 3 4 5], options);

Now the partitions are

A(:,:,1:part(1))
A(:,:,part(1)+1:part(2))
A(:,:,part(2)+1:part(3))
A(:,:,part(3)+1:part(4))
A(:,:,part(4)+1:part(5))
A(:,:,part(5+1):end)

There are definitely other ways of dividing the hyperspectral image into 6 groups, but when we are not told what properties the groups should have, we get to invent our own definitions.

[Ahmad Al Smadi]

The format of my images is .mat, the multispectral image contains 6 bands while the hyperspectral image contains 191 bands, I wanna employe the guided filter as a step of my fusion procedure. I have used ur code, it gave me this error.

Error using optimoptions (line 114) Invalid solver specified. Provide a solver name or handle (such as 'fmincon' or @fminunc). Type DOC OPTIMOPTIONS for a list of solvers.

Error in FUSION_HYPERSPECTRAL (line 25) options = optimoptions(@ga, 'Display', 'iter', 'PopulationSize', 2000);

[Walter Roberson]

It looks to me as if you must be using R2015b or earlier. Change optimoptions to gaoptimset .

[Ahmad Al Smadi]

thank you

Answer 2

[Henry Koren]

The question is what format are these images stored in? We would like to take a look at your images if possible and work on a means of decoding them in MATLAB. Would you please be able to email them to [e-mail]?

[Walter Roberson]

I think the "divide into 6 groups" part is too vague for a meaningful solution at this point. The question does not appear to be about reading the images: the question appears to be about finding the coefficients for optimal fusion of the two images, but with the meaning of "optimal" not having been given to us.

The question puts me in mind of a question about fusion of hyperspectral images that went through about 2 or 3 years ago. I do not remember much about it at the moment, but I seem to recall that it ended up being solved efficiently by using the \ operator.

Or wait... maybe I am thinking of the question involving a four band TIFF sensor image that had to be matched against a known RGB image produced by a different program, with the question being how to get MATLAB to produce the same output as the other program when the algorithm used by other program was unknown; that one went through about a year ago.