 NoelCD S L R BlissPremium,MVM
join:2003-09-03
Florida
kudos:1
4 edits | [technical] Fuzzy Reds Investigation Some of the fall photos got me to thinking... Quite often I see red objects (e.g., flowers, red shirts, taillights) in my Canon digital images that look, well, just plain fuzzy, and seem to be missing texture (this may also be true of other models). Quite often there's a serious difference in the same photo between the visual sharpness of a bright red object and the sharpness of everything else. Also, there is quite often a red fringe or fuzz around red things. You can get around this sometimes by tremendously downsizing an image, and digital sharpening can help, but why isn't the image sharp to begin with at full-size?
We start to overlook this because it's simply present in every image. However, it's actually quite bad, and figuring out how to avoid it (or fix it) would be a way to set one's images off from the crowd.
Examples:
Canon EOS 20D
50mm 1/125th F8 ISO100
Canon EOS 20D
285mm 1/640th F5.6 ISO200
Canon EOS 20D
235mm 1/640th F5 ISO200
Canon EOS 10D
17mm 1/20th F4 ISO200
I invite you to go look critically at your own bright red objects in your images.
Before beginning an investigation, I theorized that maybe...
•Red didn't focus at quite the same plane as everything else. However, that doesn't wash; expensive lenses compensate for that. Plus, bright red things NEVER seem to be sharp, at any distance.
•Perhaps more noise reduction was being done in the cameras in the red channel than others. However, since red is a component in pretty much everything that isn't deep green or blue, that didn't sound quite right either. The problem is most apparent in bright red objects.
•Is there a limitation in the JPEG compression logic that particularly affects bright red colors? But wait, really downsized images can have sharp reds, so that theory doesn't really hold water.
•Bright reds... Hmmm... Could the apparent fuzziness be a result of red channel overexposure?
•Could it just be a quirk or characteristic of Canon (and other) conversion algorithms? Maybe most programmers are using some code posted on the Internet as their base, and that code has a bug. Stranger things have happened.
I set out to try to get to the bottom of what was going on...
I looked at the exposure histograms for each of the color channels in an image I shot - which by the way had a luminance histogram on the camera that didn't touch the right side. Sure enough, in the images with the fuzzy/texture-lacking bright red objects, the red histogram had data climbing the right side of the image. Yet not in all cases. Some images with fuzzy reds had red channel data close to the right side, but not touching.
In looking at the channels in the images that weren't overexposed, one thing became clear: As the reds get pretty bright (upper-quarter of the histogram), they start to lose definition. This says the Canon demosaic/color generation algorithms may just have a small quirk - a characteristic that bright red objects lose definition.
All right, I thought, what if a different conversion algorithm were employed. Would the results be the same? I tried shooting a raw photo of red flowers and converting it in Adobe Camera Raw. What would happen? Amazingly, with some negative exposure compensation the reds came out vibrant, sharp, and full of texture.
Consider these images, the first is from the embedded JPEG Canon put in the Raw file, and the second is from the conversion through Adobe Camera Raw (Photoshop CS):
From a Canon 20D JPEG
As Converted through Adobe Photoshop CS Camera Raw
To make a long story short I've come to think that I may need to shoot Raw and use Photoshop's converter when my subject material contains bright reds. In general I prefer JPEG, but this is one area where it looks like Raw shooting may hold a real image quality advantage, for two reasons:
•Adobe seems to be able to pull more texture out of reds. I tried Canon EVU and got pretty much the same results as seen in the JPEG. Canon DPP also delivered flat reds, though they were somewhat different from the results from EVU/JPEG. Adobe seemed to do best.
•You can deal with red channel overexposure when the camera claims to have metered the shot properly (and even shows you a good histogram).
Another suggestion might be to shoot on negative exposure compensation settings whenever bright reds are present.
Sorry this was so long; just wanted to share some interesting stuff.
-Noel |
|
 dierroPremium
join:2003-10-26
East Lansing, MI
 ·Comcast
1 edit | Amazing NoelC  , it would be interesting if you got a Nikon user to post their examples using your techniques. I wonder what Canon has to say about this? Perhaps it might be something that they have overlooked, or hope their users overlook.
Again, very well written. |
|
 dandelionPremium,MVM
join:2003-04-29
Germantown, TN
kudos:4
·Comcast
| reply to NoelC
As a scientific study would be nice to see the same picture posted both ways. A big difference, is any in relation to the lighting and other factors in the picture? Also, I guess this can't be done on camera's on automatic or would this be how you save the picture? I have a Fuji, and not quite sure how to do anything with it but on automatic.
--
want to know what I'm doing? |
|
| reply to NoelC
Thank you. |
|
NoelCD S L R BlissPremium,MVM
join:2003-09-03
Florida
kudos:1 | reply to dandelion
Just to be clear, the last image with the red flowers is EXACTLY the same image done two different ways: Canon embeds a JPEG image in their Raw files. The first is the most texture in the red flowers I could pull out of the embedded JPEG, the second was what I could do by using Photoshop CS's Raw converter on the shot.
-Noel |
|
dandelionPremium,MVM
join:2003-04-29
Germantown, TN
kudos:4 | reply to NoelC
Thanks! |
|
 RubiconPremium
join:2001-02-20
Philly | reply to NoelC
NIKON D70
60mm 1/100th F7.1
|
--
R u b i c o n |
|
NoelCD S L R BlissPremium,MVM
join:2003-09-03
Florida
kudos:1
1 edit | Thanks for the Nikon image, Rubicon.
Looks like the Nikon D70 does it too. An almost unnatural lack of texture in the lower-right red leaves. Also, take a look at the fringe around the edges of the leaves. It's easier to see if you look at the red channel.
-Noel |
|
RubiconPremium
join:2001-02-20
Philly
1 edit | fuzzy reds.  |
|
 jfgnet12 Step ProgramPremium,MVM
join:2001-02-14
Limbo | reply to NoelC
NoelC,
How would that same image look with C1 conversion ?? |
|
NoelCD S L R BlissPremium,MVM
join:2003-09-03
Florida
kudos:1 | Good question. I have never gotten a copy of C1. I'd be happy to send along the raw file, though it's pretty big (7+ megabytes). I don't have a site right now with space enough to put that file online.
-Noel |
|
 HRMGod Bless AmericaPremium,MVM
join:2002-02-03
Darien, CT
kudos:1 | reply to NoelC
I believe part of the problem is the Red spectrum itself. The longer wavelength and lower frequencyu is not as good for the instrument to capture. Much like OTA channel #2 is always the worst picture on a TV due to it's bigger wavelength. There are less waves passing through the sensor in the capture than any other color.
--
I'll never forget...
FOUR MORE YEARS - Don't worry, history will be very kind to GWB.
The hype and untruths won't hurt you,
and millions of Iraqis thank you. |
|
jfgnet12 Step ProgramPremium,MVM
join:2001-02-14
Limbo | reply to NoelC
Send to jfgnet@gmail.com |
|
NoelCD S L R BlissPremium,MVM
join:2003-09-03
Florida
kudos:1 | reply to HRM
Well, that sounds like a possibility offhand, but doesn't really hold up... As I've shown, certain software delivers images without the problem. I'm thinking the entire camera industry may share some of the same basic image generation logic (and this quirk), and only of late have people implemented raw converters that don't do it that way.
-Noel |
|
|
|
NoelCD S L R BlissPremium,MVM
join:2003-09-03
Florida
kudos:1 | reply to jfgnet
Jfgnet, it's on its way. 
-Noel |
|
jfgnet12 Step ProgramPremium,MVM
join:2001-02-14
Limbo | reply to NoelC
I agree NoelC, because these Digital cameras are able to focus well in the IR range also |
|
 noe3Question EverythingPremium
join:2004-10-04
Burnet, TX
1 edit | reply to NoelC
NIKON E8700
10.7mm 1/211th F5.8 ISO50
|
Could it possibly be a monitor reproduction problem. Do the reds still look fuzzzy in a print? |
|
RubiconPremium
join:2001-02-20
Philly | reply to NoelC
NIKON D70
60mm 1/125th F8
|
--
R u b i c o n |
|
jfgnet12 Step ProgramPremium,MVM
join:2001-02-14
Limbo | reply to jfgnet
Minor problem the version of C1 I use will work on CRW files but not CR2 format |
|
RubiconPremium
join:2001-02-20
Philly | reply to NoelC
NIKON D70
116mm 1/160th -0.6ev F7.1
| 
NIKON D70
116mm 1/160th -0.6ev F7.1
100%crop |
--
R u b i c o n |
|
