Fill reduction in the Canon E-TTL II flash system is affected by Custom Function 14 on 20D DSLR.

Written 6/15/05

This document concerns a somewhat esoteric feature of the Canon Flash system called Automatic Fill Flash Reduction, AFFR. This feature is documented here so please read that if you don't already know what we're talking about here. Canon's documentation is at the bottom of the page here. It is fairly well established that under the older E-TTL flash system used by Canon, the AFFR can be disabled by means of a custom function on the camera: here are the details. Basically fill reduction means the flash will provide full flash exposure when there isn't much ambient light, but will lower flash exposure so that it just fills in shadows when there is lots of light (e.g. outdoors).

If you are new to the Canon flash system, the EOS flash FAQ is a good place to learn the basics and more advanced information.

What I wanted to determine is whether the Custom Function 14 on my Canon DSLR (which uses E-TTL II, a newer flash technology) will stop the fill reduction. My theory is that setting Custom Function 14 to 0 (Evaluative flash metering) will cause fill reduction to occur starting above about EV 10, with full fill reduction at and above EV 13, while setting Custom Function 14 to 1 (Average flash metering) will cause fill reduction to not occur. I use the notation CF14.0 to mean Custom Function 14 set to value 0, and CF14.1 means it is set to 1.

The 20D instruction manual seems to hint that CF 14 would affect fill reduction, but doesn't say it outright. Here is what it says:

C.Fn-14 E-TTL II

0: Evaluative: Fully automatic flash photography for all conditions, from low light to daylight fill flash.

1: Average: The flash is averaged for the entire area covered by the flash. Since automatic flash exposure compensation will not be executed, you may have to set it yourself depending on the scene. This also applies if you use FE lock.

The other interesting thing to explore is whether switching between manual focus (MF) and autofocus (AF) would affect the results; some had theorized that manual focus would disable auto fill reduction. I ran some tests to see whether fill reduction occurs.


Test Procedure:

I set up a scene where the main ambient light source was a 500W halogen light on a dimmer. The scene was illuminated as evenly as I could with the light close enough to produce a lot of light. The main subject is simply a bare expanse of carpet.Using a tripod-mounted Canon 20D camera with Canon 580EX flash mounted on top, I took pictures of the scene at iso 1600, f/8, in Av (Aperture-priority) mode. Inside the scene, I placed a stove pan, with some white paper inside, oriented in such a way that the paper would be illuminated as little as possible by the ambient halogen light from the side, and as much as possible by the flash. I took pictures for five different camera settings: No flash, CF14.0 AF, CF14.0 MF, CF14.1 MF, CF14.1 AF. By adjusting the dimmer on the light source, I could change the overall light level. I did these tests for EV 13.3, 13, 12, 11, 10, 9, 8, and 7, which correspond to shutter speeds from 1/160 to 1/4 second without the flash turned on. EV values can be converted to aperture/shutter values with the chart here. The pictures below are examples of the scene with no flash and with flash.



I measured the flash output by importing the pictures into Photoshop CS, converting to lab color space, and looking at the luminance info display while moving the cursor around one particular spot in the white paper inside the pan. I stored the observed pixel luminance values in an excel spreadsheet. I had shot a series of photos with different FEC (Flash Exposure Compensation) values, in order to have a calibration by which I could relate pixel values to stops of flash exposure. I interpolated pixel values between those of the third stop FEC values that I had shot. Using those values to relate the pixel values to stops of flash, I produced the following table:

Figure 1: Flash exposure produced by 20D/580EX relative to reference

EV CF14.0 AF CF14.0 MF CF14.1 AF CF14.1 MF
13.3 -0.8 -0.8 0 -0.1
13 -0.9 -0.8 -0.1 0
12 -0.8 -0.9 -0.2 -0.2
11 -0.5 -0.5 -0.3 -0.3
10 -0.4 -0.5 -0.3 -0.3
9 -0.1 -0.3 -0.4 -0.3

Because of the noise in the image and the human-directed process of selecting which pixels to sample, I would estimate that values reported are within approx .2EV of the actual captured exposures, so small differences should be ignored. Also note that zero EV in the chart above does not have any implications of "correct" exposure; I arbitrarily chose one situation for my FEC calibration, but I could have put zero EV in any other place on the chart just as easily. I am not reporting the data from EV 7 and EV 8 because I determined that the ambient light leaking through my closed curtains was increasing exposure (based on increased pixel values in the no flash shots) for EV 7-8. If anyone is interested in looking at the raw data in spreadsheet form, it is available here.


The other interesting thing is that for the low EVs in my test the camera chose an ambient exposure of one stop lower than what it chose when the flash was turned off. This phenonenem has been called Negative Evaluative Exposure Compensation (NEVEC) in a document by Musselman and Loke here. The chart of my results below shows that this lowering of ambient exposure with flash turned on starts below about EV 12 for ISO 1600, which is the same point that it occurred in the NEVEC paper referenced above.

Figure 2: Stops difference of shutter speed set by camera with flash turned on versus with flash off

EV 14.0 AF 14.0 MF 14.1 AF 14.1 MF
13.3 0 0 0 0
13 0 0 0 0
12 -0.3 -0.3 -0.3 -0.3
11 -1 -1 -1 -1
10 -1 -1 -1 -1
9 -1 -1 -1 -1
8 -1 -1 -1 -1
7 -1.3 -1 -1 -1


I assume that the -1.3 value at EV 7 for 14.0 AF in the chart above is noise in the data caused by the light dimmer and/or camera exposure system having minor fluctuations. Since I read exposures values in third stop increments, a third stop is the smallest unit of change.



This test only shows what the camera did in one particular circumstance, and cannot be generalized to all circumstances due to the complexities of the metering system, as well as the vagaries of the testing procedure. However, the data seems to indicate that fill reduction is occurring when Custom Function 14 on the 20D is set to 0 (Evaluative), and that fill reduction is not occuring (or perhaps not occuring as much) when CF 14 is set to 1 (Average).

It also seems to show that there is no difference in fill reduction between autofocus and manual focus.

Note that the amount of fill reduction is determined by a presumably complex algorithm and will presumably vary on a scene by scene basis. This test was not intended to determine the amount of fill reduction that occurs, only to determine whether it occurs at all.


How I plan to use this information:

I generally believe that using evaluative flash metering is a good thing, so I will likely continue to use CF14.0. Fill reduction is useful because it does what I would do anyway: give full flash exposure when there is not much light, and reduce flash exposure to only fill in shadows when there is lots of ambient light. However, if I am in a situation where the light is medium, then I might switch to CF14.1 if I'm going to be shooting different scenes in that light and want to control the flash exposure more tightly, to avoid the uncertainty of not knowing if a given shot will use AFFR. In any case, when shooting with flash on the 20D it is a good idea to periodically stop shooting and examine the the flash shots you have taken on the LCD, and adjust Flash Exposure Compensation if they are consistently too bright or dark.

Other people report that under E-TTL, they prefer to use CF14.1 all the time, so that they control fill reduction to avoid the uncertainty of having the camera control it. These people will likely find that the same approach works on the 20d with E-TTL II. However, the increased consistency (due to distance data and discarding questionable data) of E-TTL II versus E-TTL may convince some of these people to switch to CF14.0.


Questions about E-TTL II implementation on 20D not answered conclusively by this test:

How does ISO influence the AFFR and NEVEC under E-TTL II? Doing this test at different ISO values would be useful.

Could someone determine from their ISO, aperture, shutter, and metering settings whether AFFR and/or NEVEC will occur, and if so how?

Does the "full" amount of AFFR vary and if so, what affects it?

The EOS FAQ autofill section gives the values EV10-13 as the transition zone. Is this still true for E-TTL II?

What is the real purpose of AFFR and NEVEC? Is it successful in doing what it is supposed to do? Old Canon marketing materials say it is "to provide natural looking flash fill in most bright conditions, so that even a beginner without experience could achieve pro-quality results."


Links to related Canon Flash documents:

Chuck Westfall's archived comments on E-TTL II

NEVEC document by Musselman and Loke

EOS flash FAQ

Canon Flash Lighting Tutorial

Canon Europe E-TTL II "Fearless Flash" page

Copyright 2005 by Tom Morrow