[FFmpeg-devel] MJPG decoder picture quality
Baptiste Coudurier
baptiste.coudurier
Wed Jun 9 02:08:38 CEST 2010
On 6/8/10 4:59 PM, Pavel Pavlov wrote:
>> [...]
>>
>> This is weird, using this commandline: ffmpeg -i Tulips.jpg
>> test%d.png
>>
>> with latest svn, I don't have the same png, the png I obtain have
>> way more hot colors and looks exactly like the jpg.
>>
>> Btw the png is rgb24 not yuv444.
>>
>> Can you please share with us the ffmpeg commandline you are using
>> ?
>
>
> [Pavel Pavlov] I didn't use ffmpeg, I used avcodec. I tried doing it
> using ffmpeg -i Tulips.jpg test%d.png and I tried a downloadable
> ffmpeg build and I got the same results and in all cases images look
> the same.
Please use latest svn when doing these kind of tests.
> By the way, I didn't notice any color differences between
> my image Tulips_yuv444.png, original Tulips.jpg and test%d.png, what
> color hotness differences do you get?
The original, the png created by ffmpeg have good hot yellow color.
Your looks like the yellow was washed, which definitely suggest the
conversion from jpeg to png didn't go really well.
> Even if these images look the
> same, they really are far from the same; look closer: original image
> has perfect lines and the other two have some jagged edges and colors
> have some green or black in transition from yellow to red.
They look the same. Like Michael mentioned, the defaults a tuned toward
speed, but it can be changed of course.
> My Tulips_yuv444.png is a bit different because it's actually yuvj444p
> rendered by sdl (yuv surface) and then I simply made a screenshot of
> the window :) yuv444 is the MJPG output for hi quality jpegs, has no
> relation to png of course.
Tulips_yuv444.png is a _PNG_ using RGB24.
The original YUVJ444P was converted to RGB24 using something, ffmpeg
decoder outputs YUVJ444P.
There is nothing wrong with the _decoder_ here.
--
Baptiste COUDURIER
Key fingerprint 8D77134D20CC9220201FC5DB0AC9325C5C1ABAAA
FFmpeg maintainer http://www.ffmpeg.org
More information about the ffmpeg-devel
mailing list