An Easier UniWB for EOS Digital Cameras (at least the XT)

tdodd wrote in post #5477481
I just shot a fresh blown image with my 40D - 30" exposure at f/5.6 and 200 ISO, waving the lens in the general direction of the sun on a clear day. I then used this image to set a custom WB and the camera warned me that the image might not be suitable for WB but I went ahead anyway and then fired a sample "real world" shot. Here are the results....

Blown image
Raw Image Analyse gives RGB coefficients of 1.0000, 1.0008, 1.0015
Exif Tool give "As Shot" values of 1045, 1039, 1039, 1052

Real world image with blown custom WB
Exif Tool give "As Shot" values of 1024, 1037, 1037, 1024

That would correspond to WB coefficients of (1.000, 1.013, 1.000) so it appears to work if you force the 40D to use the image.

Thanks for the test .

e-k

On another thread tdodd brought up the possibility of setting the EXIF data directly using ExifTool to set the coefficients for WB. I had originally tried this without success in various forms and finally given up. Then I thought it must be using the image data to calculate the WB when setting the CWB. Modifying the RAW image data was a no go so I modified a JPEG and tried changing the WB Tags in conjunction with this. This was still unsuccessful.

That's when I tried the saturation/dark frame experiment which yielded positive results.

In any case, tdodd's post made me want to revisit the issue. It appears there is a tag that contains, (or is at least related to), the measured WB. This tag is unknown and so doesn't show up in ExifTool unless you ask it to display unknown tags with the -u option. If you do that you will see a tag called Canon_0x00aa with a value something like '10 640 1024 1024 965'.

Setting this value to '10 1024 1024 1024 1024' seems to work but doesn't provide a perfect UniWB when using that image for CWB. It appears the values for R and B have to be bumped up but I haven't worked out the exact value. I have no idea what the 10 is either.

Now the only problem here is that since this tag is unknown to ExifTool, you cannot use it to write to the tag. I resorted to using a hex editor.

In any case it looks promising. I was able to set the tag on a random image and use it for setting the CWB.

e-k

Okay, here's an alternative procedure. Not quite as convenient currently as one does need to resort to hex editors and such . Once again, this works on a Rebel XT and I can't speak for other models.

1. Use ExifTool -e -u to find the value of the tag called Canon_0x00aa in the image you want to use for CWB. It should be something like (10 560 1024 1024 768).
2. Convert these values to hex: (0A, 230, 400, 400, 300). In the data stream the bytes for the short are reversed so search for (0A 00 30 02 00 04 00 04 00 03).
3. Now modify the 30 02 to 00 04 and the 00 03 to 00 04 and save it.
4. Shoot another image which uses the one above as its CWB image.
5. Use the WB RGGB Levels As Shot from this new image and convert to hex like above.
6. Replace the values for tag Canon_0x00aa with the hex values from WB RGGB Levels As Shot and save it.
7. You should now be pretty close.

With the above procedure I had 1028 1029 1029 1030 initially. I modified the values a little and ended up with 1030 1029 1029 1029. Changing the R value would only oscillate between 1028 and 1030.

e-k

Hi... im new to all this and have lots of more basics thing to learn before i get into this technical stuff. But just for fun i tried the dark part of your saturation/dark experiment. And with that CWB i got almost 1.5 more Fstops before the histogram clipped than when i used a normal wb... so i guess it worked... the thing is i really don´t know exactly how to check the WB coefficients. I only have DPP. What programa do i need? it´s been fun reading about this Thanks.

ADAPTE wrote in post #5487043
Hi... im new to all this and have lots of more basics thing to learn before i get into this technical stuff. But just for fun i tried the dark part of your saturation/dark experiment. And with that CWB i got almost 1.5 more Fstops before the histogram clipped than when i used a normal wb... so i guess it worked... the thing is i really don´t know exactly how to check the WB coefficients. I only have DPP. What programa do i need? it´s been fun reading about this Thanks.

you can download DCRAW (just an .exe file runnable with no installation, download it from: http://www.insflug.org​/raw/) and try to develop any RAW that was shot with your particular WB. To do that just type in console (dcraw.exe and the RAW file being in the same folder as your console):

dcraw -v -w file.cr2

and look closely to the 4 numbers after the sign 'multipliers'. The closer they are to 1.0 the better.

There are a few different applications you can use. Here are some options,

1. ExifTool (http://www.sno.phy.que​ensu.ca/~phil/exiftool​/) will give you access to the relevant data. To check the WB coefficients use:

exiftool -WB_RGGBLevelsAsShot img_9781.cr2

replacing img_9781.cr2 with the actual file name you are checking. This will give you four values. The middle values should be exactly the same and are the green. The first is the red, and the last is the blue. You can normalise the values by dividing everything by the green value.

2. UFRaw will show the RGB coefficients if you set the WB to Camera.

3. dcRAW can be used as well. dcraw -i -v -w img_9781.cr2 will print the coefficients. If you want them normalised, drop the -i, but it will then do the actual conversion.

e-k

Thanks! i manage to use UFRaw. I just couldn`t understand the other options.

You can also use Rawnalyze. Display the RAW histogram and just above it is "Current WB Coefficients:" For a 40D shot using a CWB made with the black frame method it lists 1.000,1.000,1.000.

tzalman wrote in post #5490408
You can also use Rawnalyze. Display the RAW histogram and just above it is "Current WB Coefficients:" For a 40D shot using a CWB made with the black frame method it lists 1.000,1.000,1.000.

Hi tzalman,

I always get 1.000, 1.000, 1.000 in Rawnalyze for any picture. The apply WB is checked and greyed out. Double check if it is the same for you. If it is then the values aren't accurate.

e-k

Oops, you're right. I never looked for the WB numbers on any other image and when I saw 1/1/1 I jumped to an unjustified conclusion. I do that too much.

ISTR you have to select an area of the image first, using the right mouse click to draw a rectangle over part of the image and then press "W" or "CTRL-W" (can't remember which) to get the software to analyze the sample and calculate the respective RGB ratios.

tdodd wrote in post #5491457
ISTR you have to select an area of the image first, using the right mouse click to draw a rectangle over part of the image and then press "W" or "CTRL-W" (can't remember which) to get the software to analyze the sample and calculate the respective RGB ratios.

The CTRL-W is for calculating the coefficients based on the assumption that the area selected should be "white" (i.e. it works similar to the click for WB in some raw converters). It doesn't seem to pick up the initial WB coefficients from the RAW file -- at least for Canon images.

e-k

I dont understand whats going on here...
Can someone break it down for the slow people?

All I see are pictures turning green...

E-K wrote in post #5492099
The CTRL-W is for calculating the coefficients based on the assumption that the area selected should be "white" (i.e. it works similar to the click for WB in some raw converters). It doesn't seem to pick up the initial WB coefficients from the RAW file -- at least for Canon images.

e-k

Doh! Good point.

AdamLewis wrote in post #5492167
I dont understand whats going on here...
Can someone break it down for the slow people?

All I see are pictures turning green...

Basically UniWB is something used by people trying to expose to the right. The idea is that it will give you a more accurate view of whether something is really blown.

WB works by multiplying the various channels by a number (the coefficients being talked about). This adjusts the relative proportion of each channel.

Think of UniWB as the anti-WB. The coefficients are ideally all 1, so each channel is given the same priority.

e-k