[Image-SIG] HSL adjustment, and others
Bob Klimek
klimek at grc.nasa.gov
Tue Apr 12 16:18:45 CEST 2005
No. The paper mentioned below has the pseudocode but I've never
implemented it.
Bob
Ray Pasco wrote:
> Bob,
>
> So, that's HSI ? OK, but it seems rather bizarre. Anyhow, do you
> have the inverse transform ?
>
> Ray
>
>
> Bob Klimek wrote:
>
>> Ray Pasco wrote:
>>
>>> I've been playing around with just this. This program converts
>>> to/from several different colorspaces and changes the alternate
>>> colorspace's "ramp" slope and offset. The program only does linear
>>> adjustments.
>>>
>>> I'd appreciate any comments/suggestions you may have.
>>
>>
>>
>> Hi Ray,
>>
>> I've taken a quick look at your rgb-hsl (and hsl-rgb) and rgb-hsv
>> (and hsv-rgb) transformations and from the few tests I've run, both
>> algorithms seem to be working quite well. And both successfully
>> convert to and from. I don't see any difference really, although I
>> don't have the time to do a thorough comparison.
>>
>> My interest is only in hue so I've looked at that component only. I
>> have a test image of a type of "rainbow" in which the hue changes
>> from 1.0 down to 0.0 in a linear fashion as the image is scanned from
>> left to right. I use an old algorithm which produces a linear
>> relationship between hue and posistion in this image. To my surprise
>> both of these algorithms also produce a linear relationship.
>>
>> BTW, the algorithm that I use (see below) come from Computer Graphics
>> Quarterly Report SIGGRAPH-ACM, vol 13, Number 3, Aug. 1979. The hue
>> range is from 0 to 2pi. One interesting thing we've noticed is that
>> some hue values repeat them selves. That is an 8-bit deep RGB image
>> has 16 million rgb combinations but our hue algorithm produces only a
>> few thousand unique values, not 16 million. All others are repeated
>> hue values.
>>
>> Anyway, it would be very nice to have a fast image-wide HSI
>> transformation done in C with the output as a float array. Maybe some
>> day...
>>
>> def calcHue(self, r, g, b):
>> rf = float(r)
>> gf = float(g)
>> bf = float(b)
>> rf = rf/255.0
>> gf = gf/255.0
>> bf = bf/255.0
>> inten = (rf+gf+bf)/3.0
>> if(not((r==b) and (r==g))):
>> min = rf
>> if(min>gf): min = gf
>> if(min>bf): min = bf
>> sat=1.0-(min/inten)
>> t1 = rf-gf
>> t2 = rf-bf
>> t3 = gf-bf
>> t4 = math.sqrt((t1*t1)+(t2*t3))
>> t5 = (t1+t2)/2.0
>> t5 = t5/t4
>> hue = math.acos(t5)
>> if(gf<bf): hue=2.0*math.pi-hue
>> else:
>> hue=-1000.0
>> sat=-1.0
>> return (hue, sat)
>>
>> Bob
>>
>>
>>
>
>
>
More information about the Image-SIG
mailing list