JPEG 3-channel component ID = 0-2 YCbCr (using RGB) about twelvemonkeys

[haraldk]

Hi Boris,

Thanks for reporting!

I see your issue, and I agree, these are incorrectly recognized as RGB.

The images both has the following comment:

"Intel(R) JPEG Library, version [2.0.18.50]"

So I guess this is the software/library used to create them.

However, these images both have APP0/JFIF markers, and the JFIF spec says 3 channels should always be YCbCr (never RGB). So instead of your proposed change, I think I'll just check for the presence of the JFIF marker, and default to YCbCr for those and RGB otherwise (after testing component ids, subsampling etc). This is more generic and probably a better fix long term. Also, it works for all my test cases + extended test suite of more than 200 (more or less malformed) JPEGs. :-)

I'll try to push the fix as soon as possible, along with an updated regression test.

Best regards,

Harald K

[kendoboris]

Hi Harald,

Thanks for your quick response!

Regarding:

I think I'll just check for the presence of the JFIF marker, and default to YCbCr for those and RGB otherwise (after testing component ids, subsampling etc). This is more generic and probably a better fix long term. Also, it works for all my test cases + extended test suite of more than 200 (more or less malformed) JPEGs. :-)

That sounds great. I will be looking forward to the release :)

Thanks again.
Boris

[haraldk]

Ok, just pushed a fix for this issue:

037a47c

Best regards,

Harald K

[kendoboris]

Great! Thanks again.

Best,
Boris