I have promised to produce some test images.  Seeing is believing.

First, a PAL-sized PNG, showing a linear gradient made with Cinelerra's
gradient video effect plugin.  This PNG was rendered with the colour
format set to YUV.  It looks quite smooth, but it isn't.

Next, a PAL-sized PNG of the same gradient rendered with RGB8 as
the colour format.  This is pretty smooth.

Then a 2x enlarged region, with a 4x contrast stretch, showing
FLOAT, YUV, RGB8, FLOAT side by side.  You can see the difference
now, right? ;-)


Why is the Gradient effect so much poorer in YUV mode than in RGB8?
It's because the gradient is generated in RGB colour, then converted
to YUV (if necessary).  And the conversion is lossy, especially when
the colour depth is just 8 bits.

In FLOAT mode, the YUV<->RGB conversion is perceptually lossless,
as far as I have been able to see.  More aggressive and through
testing remains, though.  I'm sure you can squeeze out some
artifacts if you yank up the saturation, contrast or sharpness
really hard.


If you combine effects that are YUV-native and RGB-native, you can
only get decent quality in RGB FLOAT mode.  In the 8-bit modes the
image will suffer considerable loss and artifacts.

--
Herman Robak

<<attachment: gt2_gradient_only-yuv8.png>>

<<attachment: gt2_gradient_only-rgb8.png>>

<<attachment: gt2_gradient_only-stack-mid_2x.png>>

Reply via email to