Element-wise matrix division in Matlab is returning only binary values

Question

I am reading two images as matrices in Matlab and want to create a third image, which is calculated as:

(im1-im2)./(im1+im2)

However, I believe Matlab is rounding the values of the element-wise division (which of course I don't want). For example, the first value of im1-im2 is 32 and the first value of im1+im2 is 70. So, the first value of the resulting matrix is supposed to be 32/70=0.4571, but it is 0.

How can I do this matrix operation maintaining the original division result values in the final matrix?

Answer 1

It depends on your image data and how you load it. An 8-bit image loaded e.g. with imread, is stored as integer matrix (uint8). Images with higher bit-depth are stored with uint16 or more.

So you could either process your image by casting the matrix to double and cast it back to an appropriate bit-depth afterwards, or you cast it directly to your desired bit-depth and accept the rounding, as it needs to be done anyway at the end.

im1 = double( imread('fig1.png') );
im2 = double( imread('fig2.png') ); 

result = uint32( (im1-im2)./(im1+im2) );

Answer 2

im1 and im2 are uint8 or uint16 most likely, convert them to double and perform this operation. Note that converting them implicitly like *1.0 or something doesn't work.

So, something like

im1 = double(im1);
im2 = double(im2);
thingYouWant = (im1 - ...;

Note that converting back to uint will again leave you with 1 (values above 0.5) or 0 (values below 0.5). To convert back to uint8, you need to do something like:

resultInUint8 = uint8(thingYouWant * 255); % Linear scale from 0 to 255

Another thing to note is that double(im1 - im2) will result in 0 if im2 is larger than im1. So you might end up with half of image being 0. Better to convert before doing any operation.