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.
Comments
How does this compare to decoders written in Rust? From a security standpoint, rust decoders may have an edge.
https://github.com/PistonDevelopers/image-png
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.