Image colour spaces are handled inconsistently from graphics state colour spaces

[dhdaines]

Currently pdfminer.six does this for colour spaces in images (inline or XObject) when creating LTImage

        self.colorspace = stream.get_any(("CS", "ColorSpace"))
        if not isinstance(self.colorspace, list):
            self.colorspace = [self.colorspace]

This is not at all the same as the way the non-stroking colour space (which, for some reason, is the only one pdfminer.six seems to care about, and even then only barely) is handled in LTChar, where it is an actual PDFColorSpace object.

Instead, pdfminer.six should apply the same logic as get_colorspace in PDFPageInterpreter.init_resources (which could just be a standalone utility function):

        def get_colorspace(spec: object) -> Optional[PDFColorSpace]:
            if isinstance(spec, list):
                name = literal_name(spec[0])
            else:
                name = literal_name(spec)
            if name == "ICCBased" and isinstance(spec, list) and len(spec) >= 2:
                return PDFColorSpace(name, stream_value(spec[1])["N"])
            elif name == "DeviceN" and isinstance(spec, list) and len(spec) >= 2:
                return PDFColorSpace(name, len(list_value(spec[1])))
            else:
                return PREDEFINED_COLORSPACE.get(name)

Gratuitous self-promotion, this is fixed in PLAYA 😉 dhdaines/playa@96f7628