None of that explains why it only happens when converting directly to
JPEG which, if I have followed the discussion correctly, is a
characteristic of the problem.
Chuck Norcutt
Ken Norton wrote:
>> ACR would have had the problem fixed some time about June 2004 (a year
>> after the E-1 was announced). But I do find this whole thing somewhat
>> curious. If the bug is failure to interpret the RGBG pattern for
>> correct demosaicing how is it that the problem only seems to occur in
>> very bright highlights and only when converting directly to JPEG.
>> Sounds like a pretty ordinary bug to me and not something systemic
>> driven by the Bayer pattern.
>>
>
> Chuck, the way I understand it, at issue is how the nearest neighbor
> calculation is performed to determine color.
>
> The Olympus pattern is designed as a four-pixel (RGBG) quadrant whereas the
> more typical array is the triangle of red, blue and one green.
>
> You wouldn't see any difference (other than some color-space/gamut shifting
> which is another nasty byproduct of incorrect demoisiacing) in most of the
> range of colors and intensities, but once you get in the top stop of dynamic
> range, the conversion breaks down and what was intended to be a four-pixel
> quadrant with two greens no longer is able to handle the highlight
> transitions.
>
> If you work the math and attempt to manually recreate a demoisiac of the
> array you can see where it could fail in certain color and brightness
> combinations. The demoisiacing algorithm has to calculate two different
> things for each pixel position: Brightness and Color. Depending on the
> algorithm used, the assigned pixel brightness and color is determined by the
> combination of three pixels, four pixels or nine pixels. The algorithms use
> up to nine pixels to determine color for any specific pixel. If you use only
> three pixels to determine color assignment, you run the risk of color
> sparkles and maize patterns. The actual luminance value of these patterns
> and sparkles is the same as neighboring pixels, but the color is wrongly
> determined. When you increase the number of pixels used in the algorithm the
> sparkles and maize patterns usually go away.
>
> Rumor has it, and I have not been able to confirm it, so take it just as
> hearsay, that the Olympus sensor uses two different types of green pixels
> with slightly different wavelength sensitivities--not too different than the
> Sony sensor used in the DSC-F828. I would dispute that claim based on
> extensive research and detective work of the Kodak CAF sensors used in the
> E-1, E-300, E-500 and E-400. However, it would not be beyond the realm of
> possiblity that the pigments used in the Panasonic supplied Olympus sensors
> are different than the ones used in the Panasonic sensors. Doubtful, but
> possible.
>
> AG (moisiac) Schnozz
--
_________________________________________________________________
Options: http://lists.thomasclausen.net/mailman/listinfo/olympus
Archives: http://lists.thomasclausen.net/mailman/private/olympus/
Themed Olympus Photo Exhibition: http://www.tope.nl/
|