This issue is still the same with Octave 4.2.0 and image-2.6.1.

Here is a simple script to show the behavior:

clear;
pkg load image;

M = [0 1 0; 0 1 0; 0 1 0];
imrot = imrotate(M, -45, 'nearest' , 'loose', 0)

The result of current Octave (3.8.2) and the imrotate function from the image package (2.2.2 or 2.4.0) gives the following result with matrix size 4-by-4:

   0   0   0   0
   0   0   1   0
   0   1   0   0
   0   0   0   0

Matlab gives the following result, which has a different size of the output of 5-by-5:

     0     0     0     0     0
     0     0     0     1     0
     0     0     1     0     0
     0     1     0     0     0
     0     0     0     0     0

Both do use the same (default) type of interpolation method (nearest) and the same extrapolation value (0). Those two values are only explicitly given in my example script to work around bug #44794 for image-2.2.2.

Nevertheless the resulting matrix differs in size :(