I suppose from reading the article they are doing this programmatically, so it should be the code / algorithm / library used that is describing yellow as orange.
One interesting thing is - are these images encoded in the RGB color space or in a color space that allows all the colors human can see. If the latter and the code to analyze assumed RGB there may be things that look more yellow to us that would end up being interpreted as Orange. But that is a long shot.
You can go all the way through the data provenance:
- do you have all the posters for all movies? Probably not.
- do you have well-preserved examples of those posters? Many are going to be sun- or age- faded.
- do you have good scans of the posters that pre-date digital originals?
and so on.