How to reduce for loops

I would like to construct a matrix from two differnent(r,g,b) channels of a image.

by using

c_i (x,y,j)= -(‖(f_i ) ̂(Ω)‖  _2)/‖f ̂_j (Ω)‖_2

where i,j are r,g,b channels,

Ω = (m,n), neighborhood support.

code is :

%Image is I
%Ir, Ig, Ib are respective r,g,b channels of image
% code for Ir and Ig as
M=size(I,1);
N=size(I,2);
M1=ceil(sqrt(M));
N1=ceil(sqrt(N));
M2=M-M1;
N2=N-N1;
              for i=1:M1
                    for j=1:N1
                           f1=norm(Ir(i:i+M2-1,j:j+N2-1));
                        for i1=1:M1
                            for j1=1:N1
                                f2=norm(Ig(i1:i1+M2-1,j1:j1+N2-1));
                                    c((i*(M1))+j-(M1),((N1)*i1)+j1-(N1))=-(f1/f2);
                            end
                        end

                      end
                 end