Warning. This last article in this series gets a bit complex.
So far, the dynamic range tests have only altered one control at a time. In other words, the Exposure control was altered while all the other controls were held constant. Then, the Recovery control was altered while all the other controls were held constant. This approach was followed through all of the tests. This was done to get a preliminary understanding of how each control influences dynamic range and to produce some information that would be needed for the testing done in this article.
To complete an understanding of how the controls impact dynamic range, it is important to develop an understanding of two more issues. First, it is necessary to know how each control behaves on average. In other words, how does the Exposure control impact dynamic range, on average, if the other controls are tried at different settings (rather than just at the default settings)? Second, it is important to know if there are any interactions between the controls. Interactions occur when the setting of one control changes the performance of another control. For example, the impact that the Exposure control has on dynamic range might depend on whether the Medium Contrast curve or Linear tone curve is used. In other words, the Exposure control might behave differently when the Medium Contrast tone curve is used than when the Linear tone curve is used.
One approach to determining the average performance of each control, as well as the interactions, would be to try each possible combination of settings and look at the results. However, this has two problems. First, this would require that the raw file be converted at each of the possible combinations of the settings. If each of the seven controls was tried at just two settings, there would be 128 combinations - a time consuming task. Second, it would be almost impossible to look at all of those 128 conversions and figure out, in one's head, what was going on.
Luckily, there is an easier way called design of experiments (DOE). DOE is a mathematical technique (essentially a designed multiple regression) that reduces the number of conversions that are required and analyzes the results. The results are then output in both statistical and graphical format. Thus, a DOE was used to further analyze the controls' impacts on dynamic range.
For this dynamic range DOE analysis, the same raw file that was used in the previous conversions was converted sixty-six times (each at a different combination of settings). Both the measured and eyeball dynamic ranges for each conversion were determined. For the sake of simplicity, we will skip all of the statistical analysis and focus on only the graphical results. In addition, in order to avoid making things more complex, we will review only the eyeball dynamic range results. After all, as photographers, we are interested in the range of tones that can produce accurate detail. Tones that appear black to the eye are of little value -- even if they can be measured in the computer.
Each control was tested at two settings: its default setting and another setting that was selected to optimize the dynamic range. The second setting was chosen based on an analysis of the previous conversions as well as some additional tests that were done. Figure 1 shows the results of the DOE.
First, the main effects of the controls need to be analyzed. The main effect of a control is how that control performed on average. For instance, the Exposure control was tested at two settings: at the default (0) and at -1.50. The default setting was run thirty-two times. Each time, a different combination of the other settings was used. The -1.50 setting was also run thirty-two times with each run using a different combination of the other settings. The main effect of the Exposure control is how the control performed when it was averaged across all of the combinations of the other settings. Figure 2 shows a chart of the main effects of the controls.
The chart shows the main effect for each of the controls. For instance, the chart shows that when the Exposure control is changed from a setting of 0 to -1.50, the dynamic range increases slightly less than half a stop. This result is close to what was discovered in Part II of the series when the dynamic range was tested at just two combinations.
The chart shows three interesting points. First, the Blacks control has the largest effect of all of the controls. This matches what I noticed as I was running the tests. Second, the Brightness and Contrast controls have somewhat more of an effect than might have been expected from the prior tests in Part III (more about this when the interactions are examined). Lastly, all of the controls have some effect (around half a stop) and can be used to increase the dynamic range over the default settings.
The interactions are very interesting. For this article, only the first level interactions will be analyzed. First level interactions show what happens when two controls are both changed at the same time. Higher level interactions (i.e., when three or more controls are changed at the same time) will not be analyzed as higher level interactions are usually less significant than first level interactions. The best way to understand interactions is to look at the interaction graphs in Figure 3.
Figure 3 shows the dynamic range results when two controls are changed at the same time (averaged over the various settings of the other controls). For instance, the upper left box shows what happens when the Exposure and Recovery controls are both changed. The first thing that needs to be understood about this figure is that any box that has lines that are close to parallel does not have a large interaction. Thus, the Brightness/Contrast box shows lines that are pretty much parallel. So, the Brightness and Contrast controls do not have much of an interaction. This means that the two controls can be treated as independent of each other -- one of the controls can be changed without worrying about how it will affect the other control.
On the other hand, boxes that show lines that are not parallel indicate that there is an interaction between the two controls. A quick glance at the figure shows something interesting. Most of the combinations of controls either have no interaction or only a mild interaction. However, the Exposure control has interactions with several of the other controls. It is worth analyzing each of these Exposure interactions.
In the Exposure/Recovery graph (upper left corner of Figure 3), the green line shows the dynamic range with the Exposure control set at the default setting of 0. The left end of the green line shows the dynamic range with the Recovery control set at the default of 0. The right end of the green line shows the dynamic range with the Recovery control set at 33. Similarly, the black line shows the dynamic range with the Exposure Control set at -1.5. All of the other graphs function in a similar manner.
What does all this mean? Actually, this information contains a very significant lesson for photographers that want increase the dynamic range of their converted raw files beyond that which the default settings will deliver. In order to properly optimize the dynamic range of the converted image, you must understand each control and how it interacts with the other controls. Simply grabbing each control and moving it, without consideration of how it interacts with the other controls, could actually result in either no increase or a decrease in dynamic range.
I am sure that some of the readers are thinking, "Enough with the main effects and interaction stuff, what is the best combination of settings to optimize the dynamic range?" As with so many of my articles, the answer is, "It depends". A look back at Figure 1 shows that there are several combinations of settings that will result in a fairly optimized dynamic range (i.e. nine stops or greater for this camera). It all depends on what one wants to achieve. If one wants to recover highlights, starting by decreasing the Exposure control would be a good start. If one wants to recover shadows, starting by decreasing the Blacks control would be a good idea. The other controls can then be adjusted to finesse the results. Some experimentation might be required to create the best results for any particular file.
These results were produced with a Canon 5D at 100 ISO. My belief is that the general trends are the same for other DSLRs (e.g., decreasing the Exposure will recover highlight detail). However, there will be some degree of variation from camera to camera. The best way to discover the exact impact of the controls with your camera is to experiment with your files to see what results you get at different settings.
If one wishes to extract every bit of dynamic range from a file, there is another approach that may be very effective. This approach requires two conversions to be made from the same file. One conversion is made that extracts the detail from the highlights. The other conversion is made that extracts the detail from the shadows. The two conversion must them be combined in the computer (for more detail on one way to combine the conversions, see Digital Graduated Neutral Density Filter).
All of this main effects and interaction information may seem like way too much data to understand. After all, we took up photography to create great images -- not to run DOEs and process the data. So, what should one do? For the average photographer, an understanding of how the Exposure, Recovery, and Black's controls impact the dynamic range is a good start. These controls will allow one to increase the dynamic range over what the default settings produce. Alternatively, one can use the two conversion approach. Those photographers that wish to have the maximum amount of control over the dynamic range will need to learn all of the controls and how they interact.
Our cameras are capable of producing greater dynamic ranges that the default settings in Camera Raw produce. A little bit of understanding can help us extract greater dynamic ranges during the conversion process.