Why does legacy code scale variables by 2^16?

I am trying to understand some legacy Matlab code, and was puzzled to note that many variables are multiplied or divided by 2^16. After some digging I found this is the largest number that a 16-bit (unsigned) integer can hold, see here. I believe the code originated in C or C#. I have two questions: (1) Could the code author be using this factor to force the variable to be a float? (2) If so, can I delete all the 2^16 factors without affecting the values in Matlab?

Here is an example,

SOME_THRESHOLD = floor(0.010 * 65536); % Author's comment indicates this is supposed to represent 10%

SOME_THRESHOLD gets passed in to a function which where all the other variables have been multiplied by either 65536 or 2^16 prior to arithmetic operations; floor and abs are also used.

[Rik]

This really depends on the code itself. Usually Matlab will be backwards compatible, but that really depends on the code itself.

[KAE]

Your answer refers to question (2) above, right? I apologize for putting two questions in the same writeup, and edited my question to make this clearer.

[Walter Roberson]

It might be a storage matter: the authors might have stored as 16 bit but rescale for actual calculation.

Also, if the code is doing image processing work, it is common to rescale the pixel values to the range 0 to 1 before doing the calculation, as doing so can cut down on intermediate rescaling and overflow checks. Some of the algorithms having to do with colorspace conversion are defined in terms of the range 0 to 1.

[Image Analyst]

Care to share the code so we can comment more precisely? Or else ask the author why they did it.

[KAE]

The author is unknown, unfortunately. I will add an example to my question but honestly it's just multiplication or division of some of the variables by 2^16 or 65536. I found this writeup that makes me think it is a leftover from the original C code, and the technique dates from the 1980s!

[Image Analyst]

I don't see the example in your original/edited question. But ... good luck though.

[Christoph F.]

I think the original author tried to make the MatLAB script use exactly the same numbers as the C code would.

Answer 1

[KAE]

Just to close this question out, based on all the comments and info at the links, it appears that the 2^16 factors are for binary scaling in the original C code. There could be numerical differences if the 2^16 factors were removed but they are small for my application. Thanks for all your help!