Amount of decimal values after data normalisation

Hi guys,

I hope I can explain myself in a way that help you understand my issue. I'm working with the image processing toolbox. I have images in different bands that I use as input to feed a neural network outside Matlab. The neural network is also fed with a series of values as pixel classes, in order to segment the images. The data classes are collected with a spectrometer and then normalise to 0-1, in which case it gives something like 840.413 (wavelength) and the intensity value 0.6587563. Before running the algorithm, I check the images individually in Matlab using the image processing toolbox. The images are aligned and radiometrically corrected using field calibration panels. The original images are loaded as 0-255, then I use either mat2gray or im2double to normalise them as 0-1. My problem is that I end up having more pixels identified within the range I'm looking for than I should have. Is it possible that giving a certain input with 7 decimals does not make any difference, since the normalisation has just two decimals? Clearly, 0.65 is not the same as 0.6587563 and therefore, I might be using a value that is way over the normalisation that mat2gray or im2double do?

[dpb]

I don't understand where the "since the normalisation has just two decimals" thought comes from? Either returns a full-precision double scaling the input from its min/max to 0-1.

NB: If the input image is uint8() there are only 256 levels possible so there will only be 256 maximum possible different output values as doubles as well.