You question is very unclear due to imprecise language.
Are you intending to binarise the image? That is convert each pixel to a single 0 or 1, as you've done with your b = round(a / 256), which convert all pixels below intensity 128 to 0 and all above 127 to 1. Or as your watermarking tag and question title suggest, convert each pixel to a sequence of bits?
If the latter, the best course is actually not to do it and use matlab's bit wise operators. This will be much faster than converting to a matrix of bits and back.
Example, replace LSB of matrix with random 0 or 1:
orig_image = imread('cameraman.tif'); %demo image shipped with matlab
lsb_replacement = uint8(randi([0 1], size(orig_matrix))); %random replacement
%set lsb to 0 (with AND 254) and replace with OR:
new_image = bitor(bitand(orig_matrix, 254), lsb_replacement);
imshowpair(orig_image, new_image, 'montage')
isequal(bitget(new_image, 1), lsb_replacement) %should return true
