whos -file imageSample.mat
load imageSample.mat
image(PerfImg)
We don't know if that is what would be expected or not; note that the given array as uploaded is not int32 as far as data type but an array of double(), the default MATLAB data type.
MATLAB supports the 8- and 16-bit depth as common image format types but doesn't have a specific format for int32; hence the other routines that expect such don't work.
[min(PerfImg(:)) max(PerfImg(:))]
It appears you may have scaled the above to 8-bit magnitude based on the above range, but kept the full precision of the double?
Why not attach one of the original images and let one of the real image processing gurus see what they might think?
The disadvantage of using double() is simply one of storage; each pixel is 8-bytes; int32 is half although then overall storage also depends upon whether the original has more than the one plane or not. Of course, the size of the sample, anyway, is small enough that storage shouldn't be an issue regardless of data type chosen.
A secondary issue is what the end result is to be -- if need to save as an image file that other applications would be able to open, then see imwrite to see what MATLAB supports in that direction.
