If you look at that crate's issue tracker you can see open issues titled "Crash on malformed input", "4-bit grayscale image is read as 8-bit", I don't think it's safer or even tested to be correct. For spng there is continuous regression testing which verifies the decoded images against libpng and is fuzzed by OSS-Fuzz (https://github.com/google/oss-fuzz) just like libpng.
Curious why, seeing as the source code is so small, you don't just make the single source/header pair the default (and only) distribution? Why even split it into separate files?
https://github.com/PistonDevelopers/image-png