double precision becomes complex double after calculation

The outputs for a becomes complex double after calculation. However, it should be double precision since I am just raising the power of the element. Is there any possibility to fix it? .

i got a as a comlex double and in b i got NaN values.

tic
clear;
close all;
I=imread('Lenna.tiff');

if size(I,3)==3
    I=rgb2gray(I);
end
k=I;
%size of image
[M,N]=size(I);
%convert to double
I=double(I);
Y=zeros(M,N);

binimg=de2bi(I,'left-msb');
binimg1 = binimg(:, 1:end-3);
binimg2 = bi2de(binimg1,'left-msb');
finalimg = reshape(binimg2,size(I));
I=finalimg;
% ----------------------------------------------------
%% Block Truncation Coding Compression
blksize=2;    %Block Size
mu=colfilt(I,[blksize,blksize],'distinct',@(x) ones(blksize^2,1)*mean(x));
sigma=colfilt(I,[blksize,blksize],'distinct',@(x) ones(blksize^2,1)*std(x));
q=I>mu;
q=colfilt(q,[blksize,blksize],'distinct',@(x) ones(blksize^2,1)*sum(x));
m=blksize^2;                          %length*width of block
a=mu-sigma.*(sqrt(q./m-q));           %low mean
b=mu+sigma.*(sqrt(m-q./q));           %high mean
H=I>=mu;                              %elements of Bitmap
Y(H)=a(H);
Y(~H)=b(~H);
Y=uint8(Y);   
% ----------------------------------------------------------------------------
% Dividing into blocks
% ----------------------------------------------------------------------------
if ismatrix(Y)
  blocks2x2 = mat2cell(Y, 2 * ones(1,256), 2 * ones(1,256) );
else
  blocks2x2 = mat2cell(Y, 2 * ones(1,256), 2*ones(1,256), size(Y,3) );
end
% -----------------------------------------------------------------------------
subplot(1,2,1),imshow(k);title('orginal image');
subplot(1,2,2),imshow(Y);title('compressed image');
%output BTC image
% dPSNR=psnrr(I,Y);
toc

[Walter Roberson]

your q looks like it will be nonnegative (possibly 0). Your low calculations have sqrt(q./m-q) . The inner expression factors to q.*(1/m-1) . With m being positive the 1/m-1 is negative . But q is nonnegative . Nonnegative times negative is nonpositive. Therefore you are taking sqrt of something that is either 0 or negative .

In your calculations of b notice that q./q is going to be 1 except when q is 0 so it is not obvious that q./q is appropriate there .

[niranjan v]

i got it sir thank you so much.

can you provide the modified code.

[Walter Roberson]

No, I cannot provide the modified code as I do not know the proper equations.