Introduction to the non-technical filmmaker
As any Cinematographer knows, a camera has a limited range of exposure. This means that the amount that the camera can ‘see’ between the a very bright object and a very dark object, is limited. A common example for this is a person standing in a room next to a bright window. Any modern camera would have issues with this high contrast situation. The Cinematographer would have to make certain choices; Do we want to see outside the window and keep the person standing as a silhouette? Do we want to see the person and give up on the bright window? Do we need to light the person, and if so – how much?
These are merely some of the solutions to this problem. No matter what the chosen approach is, someone needs to make an informed decision. This camera test takes on a technical approach (what we call a quantitative test) to figuring out – how bright is ‘too bright’? How dark is ‘too dark’? It will help in the design of future camera tests and ultimately decisions made by the Cinematographer.
To the left is an example of a situation where bright objects in the background are in danger of ‘disappearing’, and in fact some do. One cannot rely on the monitor on set to make informed exposure decisions in a timely manner. It is a result of knowledge and experience gained over time.
This test was executed to allow the evaluation of the dynamic range and latitude of the Arri Alexa Mini. Questions pertaining to the effect of resolution and internal 4K UHD upscaling on noise and latitude should also be answered, as well as signal to noise level comparison of different exposures. Finally, the effect of changing the camera EI on the distribution of usable stops will be observed. All tests performed for both Tungsten and daylight color temperature setting.
In the past, film negative was evaluated by using a gray card and a densitometer. Since exposure is directly linked to the density of the metallic silver crystals, it is possible to measure it and create a sensitometric curve which characterizes the response of the negative to exposure changes. This logic was taken and applied to create a ‘digital density’ test.
In this test a gray card is filmed at different exposures, covering the dynamic range of the camera. The ‘density’ is found by measuring the mean (average) value of the pixels in the final image. The difference between the pixels represents noise, it is used as a base to calculate the signal to noise (S/N) ratio. All files evaluated were 12bit DPX files. Changes in exposure we made in such a way to have the least undesired side-effect on the final image (lighting adjustments we made with the generous assistance of Cinematographer Hana Kitasei).
A qualitative component is added, since latitude is defined as the permissible change of exposure which does not affect image quality. In this case, a human face (with the generous assistance of Cinematographer Lauren Guiteras) will be the reference for quality. It is easier to detect quality changes on human skin tone as we are instinctively sensitive to these changes.
This test was recorded in ProRes 4444 XQ in LogC.
Other settings, such as white balance, resolution, EI, shutter angle etc. were changed as needed. The lens used was Canon 15.5-47mm. The aperture was kept above T4 for the majority of exposures to prevent the effect of vignetting on the results.
Problems with the test
Light meters are not born equal. Each will give a slightly different result than the other. Therefore one needs to take this slight deviation into account when using these test results. They can still be used to find a great deal of information about this camera and it’s performance, as well as creative decisions when used properly.
The Digital Density curve represents the response of the camera to changes in exposure. The vertical axis shows pixel values on a 12bit scale, 0-4095. The horizontal axis represents changes in exposure. N, or ‘Normal’ represents the reference exposure where the spot meter reading on the gray card matched the aperture setting on the lens.
Three color traces are created, one for each color channel. They are difficult to separate at times, which indicates no specific color shift. The gray trace indicates a reference curve for rec709. The LogC image was used with a standard rec709 LUT and the luma channel was extracted to create this trace. These types of LUTs are rarely used alone, but it is a good reference for the expected final contrast, or the amount of information that may ‘survive’ in the image after color grading.
The S/N curve represents the changes in final image noise as exposure is changed. The ratio between the mean value to the standard deviation of the ‘background’ (Dmin) is part of the calculation here. Overall, a value lower than 30db indicates noise that is visible and possibly best to avoid in important image areas – although this is a subjective assessment which should be made by the artist.
The graphs below represent all the results from 10 individual tests performed. Press on any image to enlarge.