convert int16 to uint16 without losing information

I have a range of int16 values from -32768 to 32767 and want to convert these to uint16 values from 0 to 65535. I can't get this to work. Negative values get truncated to zero.

[Adam]

Can't you just cast to double, add 32768, then cast to uint16?

Answer 1

[Bob Read]

Thanks to the person who suggesting casting to double, that's solved my problem.

[Adam]

If this is something you do more than once I'd recommend creating a function to do it though so you don't need to do the maths each time. There may be a function in Matlab that does this, but neither cast nor typecast do.

There are functions in the image processing toolbox that do some kind of more friendly casting though, I just can't remember what they are off-hand. They'd be no better than writing your own as above though.

[Guillaume]

The image processing functions are im2xxx, in this case, im2uint16. They're implemented as mex files and hopefully only involve bit twiddling so should be faster than converting to double and back (which is going to be slow relatively speaking).

Answer 2

[nicolas potier]

https://la.mathworks.com/help/matlab/ref/typecast.html?searchHighlight=typecast&s_tid=doc_srchtitle

Esta es la función que necesita. Espero le sirva - >> typecast ----->Guarda los datos y no los pierde

[Adam]

typecast won't shift data to the new range, it will just wrap the data within the new data type e.g.

>> typecast( int16( -32768 ), 'uint16' )

ans =

uint16

   32768

[Guillaume]

At that point, you only need to toggle the MSB and you're done.

Answer 3

[Guillaume]

It's an old thread but since it's been revived: In theory, the following should be faster than transitioning to double as it's just very simple bit toggling:

testarray = [intmin('int16'), -1, 0, 1, intmax('int16')]
asuint16 = bitxor(typecast(testarray, 'uint16'), uint16(32768))

In practice it may be the same speed, depending on how uint16 is implemented. If it read 32768 as double then convert to uint16, then it'll be slow (As conversions between floating point and integers are a lot more involved than just bit twiddling)