The png format only supports integer pixels (up to 16 bits). There are actually very few mainstream image formats that support floating point pixels as for visualisation the decimal part makes no difference.
In theory, you could save the image as png (converting it to integer) while storing the original image in a custom png chunk, but matlab does not give that level of control over load/save of png images.
I'm sure we've told you this before: if you want to save the image for visualisation, convert it to integer, you can use uint16 instead of uint8 for more precision. If you want to save the image for further precision save it as mat file. If you want to do both, save it in both formats.
