4.1 Charts
4.1 Charts
OECF
Opto-Electric Conversion Function
The goal of the OECF is to map the scene exposure to the code value. This was done using the OECF chart and the color sample tool in Photoshop. The color sampler tool gives normalized values that have been normalized on a 10-bit scale, thus permitting code values higher than 1 for files such as ACES files, which are 16 bit.
The Film OECF For Original Printing Density And ACES
Fig. 7-1. Film printing density off Arri scanner
N for 500T at -1.7
Fig. 7-2. Film after ACES transform
N (at 0.02 lux-sec) This is not normalized to 10bit
From the printing density graph above, there are over 9 stops of dynamic range and we have not even hit the shoulder of the OECF curve yet for film. From the ACES OECF plot, we can see that it has linearized the film printing density and expanded into 16-bit float space as there are code values above the 10-bit normalization of 1.
The reason for the film lifting off of the Dmin in the overexposure was due to flaring when the lens was wide open, which causes the overall black exposure to be lifted off of the Dmin.
The D21 has also been processed through this same method to compare the logC space to what has been done by the ACES IDT. Below are the OECF charts for the D21 logC and ACES spaces:
The Film OECF For Original Printing Density And ACES
Fig. 8-1. D21 logC before transform
Fig. 8-2. D21 after ACES transform
The D21 shows about 8 stops of dynamic range in the graph above, but similar to the film camera, the lenses were not fast enough to provide enough overexposure to find the shoulder of the OECF curve. The ACES transform appears to have done the same thing, but there is a slight noticeable curve to the ACES OECF plot. This may be due to the IDT not being perfectly calibrated to the D21, causing a slight shift in the linearization on the D21.
Noise
To calculate noise, any uniform expanse can be used. We took test footage of a gray card at a normal exposure, two stops under, and two stops over to test the noisiness of each camera system.
To calculate noise from a digital image, the process is similar to how a densitometer would work for film. Samples are taken of the image at a perceptually relevant size and the standard deviation of all these samples relates to the noise.
First, the gray square of the gray card was isolated from the image.
Then, square samples of the image were gathered, the size determined by equation (1)
Where the magnification relevant to the human eye is expressed as: equation (2)
Equation (1)
Equation (2)
Assuming the original media is 11.8 mm in the vertical dimension, the height of a 35 mm film frame, and the image is being viewed 3 picture heights away, then the aperture diameter should be 56 μm.
The D21 images are 2160 pixels tall, so this translates to a sampling area of approximately 11x11 pixels. For our film, the sampling area is about 15x15 pixels for a picture 3112 pixels tall.
These samples were taken across the image, so that no samples overlapped. The mean of each block was calculated and the standard deviation of all the samples were found.
This was performed for each color channel individually, and then the overall noise was found by taking a weighted sum: 0.6 of the green, 0.3 of the red, and 0.1 of the blue channel
Table IV Gray Card Noise (16-bit CV)*
These numbers confirm what was visibly apparent from the images: the film images are considerably noisier than the D21 images.
For the film, the noise gets worse at underexposures, and better at overexposures.
For the D21, it is the opposite. The digital camera actually gets slightly less noisy when underexposed, except in the blue channel.
It is expected that a higher exposure would yield less noise, as there is more light available, which tends to even out the exposure.
Another trend to note is that for the film, most of the noise comes from the blue channel. In fact, the red and green channels are even better than the D21 in some cases. However, the blue channel is so noisy it makes the rest of the image look much worse.
This is important to remember for visual effects and the tracking scene
It is useful to have a measurement of noise based on perception as well. This is measured in grain units (GU), in which one GU is approximately one just-noticeable difference in noise. (3)
Equation (3)
Performing this calculation for each of the exposures gives 5.02 GU for a normal exposure, 8.37 GU for the underexposure, and 0.36 GU for the overexposure. This means that the noise difference is imperceptible for overexposed images, but for underexposed ones, the difference is extremely obvious.
Macbeth Color Checker Chart
The MacBeth Color Checker Chart was analyzed to determine the difference in colorimetry between the two camera systems. Because we want to test the raw capabilities of each camera, no color correction was performed before taking measurements. The raw images used to perform these calculations are shown below:
Average L*a*b* values for each color patch were extracted from the images in Photoshop.
To find the same values for each patch before reaching the cameras:
XYZ values for each patch are calculated using the spectral reflectances and
assuming a D65 white point. (4)
Which gives the Y and Z values as well by swapping in the appropriate color matching function in the numerator.
L*a*b* values are calculated using the following equations: (5) (6) (7)
Equation (4)
Equation (6)
Equation (5)
Equation (7)
Where: (8)
Equation (8)
Then, the ΔEab color differences were calculated for both cameras. (9)
Equation (9)
These calculations yielded the following results for each of the cameras:
From these results:
Table V Eab
It is clear that neither camera’s color reproduction is very accurate to colorimetry.
This is acceptable though, as cameras are not designed to reproduce colors exactly, but rather to produce an aesthetically pleasing interpretation of that color. However, the colors should be similar enough that they are recognizable.
The film’s color reproduction is overall more accurate than the D21’s. Some of this is due to the encoding scheme of the D21.
The LogC curve produces a very low-contrast image, but this is intended to be expanded in post, so the raw image from the camera is not necessarily relevant to how the colors will actually appear in the final image.
This encoding method is not intended to give a pleasing image because it is assumed that the image will be digitally manipulated. Instead, it is intended to preserve as much information as possible.
However, the D21 also had a magenta bias in all images, which certainly impacts the perceived colors in the final result. The film, on the other hand, produces a fairly pleasing image before any post-production work.
Some color correction is still desired to get the best color possible out of the image, but without this work, the raw images still look very good.
Notes
These measurements from the Noise table are in Grain Units.
