A Brief Analysis of the EOS 350D/RebelXT Histogram and Highlight Warning Feature

Recently I've been looking into finding an easier mechanism to determine a UniWB image for use in certain Canon cameras. Truth be told, it was more out of interest and fun than any pressing need. Most of my current images have been in dark dungeons were I'm pressed to get even a proper exposure let alone being able to ETTR .

In any case, it raised my curiosity about the efficacy of the 350D's histogram and highlight warning feature.

To test this I took an existing JPEG image produced by the 350D and modified it using a photo editing program (the GIMP). The first modified image was used to determine the gray point which activated the highlight warning. I continually modified the gray level until cameras highlight warning started blinking. Using this method the activation point was determined to be about (249, 249, 249). This allowed an approximate luminance value of 249 to be established for the activation point.

The next step was to determine the weighted balance of the channels. To do this, I used a similar method to above but changed 2 of the channels to different but constant values while varying the 3rd channel. The 3rd channel was varied until the highlight warning activation point was determined.

Several data points were established using the above mechanism. Now, since it was a JPEG image there were slight variations in the channels when saving so the procedure isn't exact and more data points would have been nice.

I acquired 4 data sets in this manner. I then proceeded to perform a least squares on the data and finally solved the set of linear equations to provide an approximate answer of (0.29, 0.58, 0.13). This is a little off but reasonably consistent with the Cambridge In Colour article on histogramswhere values of (0.30, 0.59, 0.11) are stated.

You can see this somewhat by producing a JPEG image which consists solely of three bars, one each of red, green and blue. When you look at the histogram you will see three sharp spikes: 2 in the left half and one which is around (a little right) of the halfway point.

So what does this mean in practice? In general it means that there are going to be situations where clipping will not be indicated despite one or more channels being blown.

Take for example a very blue sky and a very green pasture. On the top portion, even if you reached saturation on the blue channel, this would be reported as say (120, 120, 255) in the JPEG data. This would correspond to a luminosity value of about 135 which is not enough to activate the highlight alert. Same would apply to the green channel. Your histogram would also look like it hadn't completely reached the right side and that you had room to increase the exposure.

e-k

Interesting info... how does that tie into Uni-WB, though? You would still be able to see you were close to clipping in an RGB histogram, correct?

Yes, in an RGB histogram you should be able to see if you are close to clipping and UniWB is very useful in that context. With the luminosity only histogram, which is all I have access to with the RebelXT, UniWB can be problematic.

In general, I think ETTR is more problematic with a luminosity only histogram and a highlight warning feature implemented like the 350D. This is because you cannot just depend on the blinkies or histogram to see if you are clipping.

I should add though that the experiments were with extreme values. In real life, it is less likely that you will have these extremes.

e-k

My understanding is that the luminance histogram matches the green more closely than red or blue - would it be worth skewing the green channel so that it more closely matches the overexposure warning (if that's possible)?

mnealtx wrote in post #5544686
My understanding is that the luminance histogram matches the green more closely than red or blue - would it be worth skewing the green channel so that it more closely matches the overexposure warning (if that's possible)?

Green accounts for approximately 60% of the luminance value which is why it follows the green channel more closely than the others.

The overexposure warning and histogram will match as they are both based off of the luminance value. So skewing one will skew the other as well.

The way WB generally works is that you multiply the red and blue channels by a multiplier whilst the green remains constant. With UniWB, the red and blue channels are not adjusted (or at least not by much).

I need to do some more real life tests, but I suspect there may be some circumstances were it makes more sense to use a non-UniWB setting to ensure that the green channel doesn't get blown.

e-k

E-K wrote in post #5543336
Yes, in an RGB histogram you should be able to see if you are close to clipping and UniWB is very useful in that context. With the luminosity only histogram, which is all I have access to with the RebelXT, UniWB can be problematic.

I tend to see it differently. With an RGB histogram there is less need for UniWB because in most lighting situations one need only watch the green channel because if it is unclipped the chances are very good that the red and blue will not be clipped in the RAW. The exceptions are when the light is strongly deficient in the central (green) part of the spectrum, e.g. light reflected from very saturated red or blue objects like flowers or plastic toys. In such cases one would be well advised to watch the appropriate histogram. However, as I said, for most scenes the green histogram should be a reliable guide and the primary value of the UniWB is not for creating better histograms (since the green remains essentially unchanged) but rather in preventing false "blinkies".

With only a Luminousity histogram and UniWB, however, because the red and blue values are not inflated, they have even less influence in determining the shape of the histogram, i.e., it resembles even more the green channel and can, in most cases, be trusted as one would trust a green only histogram. And, of course, in those same circumstances where the green alone is not adequate the UniWB/Luminousity histogram would be misleading.

The bottom line is that when using this tool, as when using the camera in general, the successful photographer needs to analyze the scene before him and understand how his camera will handle the incoming data.

tzalman wrote in post #5546466
...The bottom line is that when using this tool, as when using the camera in general, the successful photographer needs to analyze the scene before him and understand how his camera will handle the incoming data.

I agree completely which is why I performed the analysis . I felt it was important to understand how both my histogram and the "blinkies" were implemented in my camera.

All I'm saying is that based on this analysis there is the possibility that by using UniWB with the 350D you run the risk of accidentally blowing the green channel.

e-k