The approach I take is to pre-allocate the final image, then just "drop in" the image data in the appropriate locations. This works for regular tiling, including translation. I've never had occasion to apply a rotation, but hopefully this is applicable. Perhaps use imrotate on the tiles before trying to add...
fullmask(ttop : tbottom , tleft : tright , :) = imadd(fullmask(ttop : tbottom , tleft : tright , :), temp_img) ;
Where fullmask is your final image, and ttop, tbottom, tleft, and tright are the coordinate extremes of the image to be added. Such that if you had a 100x100 pixel image to add in the upper left corner, ttop = 1, tbottom = 101, etc.
