linearRGB conversion to/from HSL
Question
Does anyone know of a way to get HSL from an linearRGB color (not an sRGB color)? I've seen a lot of sRGB<->HSL conversions, but nothing for linearRGB<->HSL. Not sure if it is fundementally the same conversion with minor tweaks, but I'd appreciate any insight someone may have on this.
Linear RGB is not the same as linearizing sRGB (which is taking [0,255] and making it [0,1]). Linear RGB transformation from/to sRGB is at http://en.wikipedia.org/wiki/SRGB. In VBA, this would be expressed (taking in linearized sRGB values [0,1]):
Public Function sRGB_to_linearRGB(value As Double)
If value < 0# Then
sRGB_to_linearRGB = 0#
Exit Function
End If
If value <= 0.04045 Then
sRGB_to_linearRGB = value / 12.92
Exit Function
End If
If value <= 1# Then
sRGB_to_linearRGB = ((value + 0.055) / 1.055) ^ 2.4
Exit Function
End If
sRGB_to_linearRGB = 1#
End Function
Public Function linearRGB_to_sRGB(value As Double)
If value < 0# Then
linearRGB_to_sRGB = 0#
Exit Function
End If
If value <= 0.0031308 Then
linearRGB_to_sRGB = value * 12.92
Exit Function
End If
If value < 1# Then
linearRGB_to_sRGB = 1.055 * (value ^ (1# / 2.4)) - 0.055
Exit Function
End If
linearRGB_to_sRGB = 1#
End Function
I have tried sending in Linear RGB values to standard RGB_to_HSL routines and back out from HSL_to_RGB, but it does not work. Maybe because current HSL<->RGB algorithms account for gamma correction and Linear RGB is not gamma corrected - I don't know exactly. I have seen almost no references that this can be done, except for two:
- A reference on http://en.wikipedia.org/wiki/HSL_and_HSV#cite_note-9 (numbered item 10).
- A reference on an open source project Grafx2 @ http://code.google.com/p/grafx2/issues/detail?id=63#c22 in which the contributor states that he has done Linear RGB <-> HSL conversion and provides some C code in an attachment to his comment in a .diff file (which I can't really read :( )
My intent is to:
- send from sRGB (for example,
FF99FF
(R=255, G=153, B=255
)) to Linear RGB (R=1.0, G=0.318546778125092, B=1.0
)- using the code above (for example,
the G=153 would be obtained in Linear
RGB from
sRGB_to_linearRGB(153 / 255)
)
- using the code above (for example,
the G=153 would be obtained in Linear
RGB from
- to HSL
- modify/modulate the Saturation by 350%
- send back from HSL->Linear
RGB->sRGB, the result would be
FF19FF
(R=255, G=25, B=255
).
Using available functions from .NET, such as .getHue
from a System.Drawing.Color
does not work in any sRGB space above 100% modulation of any HSL value, hence the need for Linear RGB to be sent in instead of sRGB.
Solution
It doesn't make much sense to convert to linear RGB, since HSL is defined in terms of gamma encoded values. Instead, write your own function convert sRGB to HSL, modulate the saturation with those values (allowing potentially out-of-gamut saturation values), and then convert back to sRGB, clamping intensities that are out of sRGB range (or disallowing saturation changes that can't be encoded in sRGB).
OTHER TIPS
The System.Windows.Media.Color class lets you get or set scRGB via ScA,ScR,ScG,ScB properties, or RGB via A,R,G,B properties.
So you could convert RGB to HSL, manipulate that, then convert back to RGB and store in a Color instance. You can then read out the converted scRGB properties.
Not ideal, and might involve some information loss. But it's an option!
Based on your comment here, your issue isn't doing the conversion incorrectly; it's that you are performing successive conversions on quantized values. In other words, you are taking the result of sRGB=>RGB=>HSL=>HSL=>RGB=>sRGB and using that in subsequent color operations. The most straightforward way to maintain precision is to always keep around the original RGB value and accumulate the changes in HSL space you want to apply. That way, you are always applying the HSL space operations to the original color and you don't have to worry about repeatedly processing quantized values.
Does this help? There are a lot of interesting links in that question, maybe something that works in your case, too...