Drew Moe - Cinematographer

Digital Camera Noise Tests

Modern digital camera sensors have become incredibly sensitive to light, affording the Cinematographer an unprecedented ability to shoot in low-light or no-light scenarios. Along with the advent of a new generation of low-energy, low-heat, compact lighting technologies (LED, Plasma, etc), the way a modern film set is lit and exposed has shifted dramatically. With this increase in light-sensitivity, however, also comes the increased risk of digital noise — an inevitable side effect of pushing a digital sensor to its limits. Exposing the sensor in a low-light scenario, therefore, becomes a careful balancing act between acceptable noise and needed sensitivity.

^{ARRIRAW at 200 ISO}

Testing Methods

Archived below is a series of test images of an MSCCC Color Checker Classic color chart, an 18% Grey Card, and a Neat Video Noise Profile Target. The charts have been shot with flat, soft lighting. Exposure and white balance have been normalized. The charts have been shot at various ISO levels, with shutter speed, f-stop, and/or lighting adjusted to bring exposure back to normal.

To quantify the noise characteristics of each camera, I've sampled a 200 × 200 pixel portion of out-of-focus 18% grey. I then refer to a histogram of the sample and its standard deviation (SD) from the mean. A perfectly noise-less sample would have an SD of 0, since every pixel would have the same value. This does not account for differences in resolution, however, as the noise within a 200 × 200 pixel sample will constitute a smaller portion of a higher-resolution image versus a lower-resolution image.

^{A standard deviation of 3.38 in the red channel}

I also captured a noise calibration target, to be used with the OpenFX Neat Video noise reduction plug-in. I've provided download links to the noise profiles derived from these tests. If you have the plug-in, you can apply the appropriate profile based on your camera and ISO settings to eliminate unwanted noise. You can learn more about the Neat Video plug-in here.

All images have been exported as JPEG files at full quality and 1920x1080 resolution.

Click on any thumbnail for a larger view.

Due to the conversion into JPEG files for web-viewing, the images presented should only be considered a rough reference. The standard deviation calculations and downloadable noise profiles, however, have been taken directly from the original camera data, before any JPEG conversion.

Wherever neccessary, the original camera data has been converted from the original raw/log color space and into a standard REC709 color space. All images and calculations are from that REC709 reference, not the original raw/log image.

Results

Arri Alexa XR (ARRIRAW)

Tests were done using an Alexa XR, recording ARRIRAW at 2.8k resolution in LogC. Exposure and color was normalized across all shots. A LogC to Rec709 LUT was applied, from which 16-bit DPX files were created. All calculations were taken from these DPX files. Presented here are JPEG images, exported at full quality and 1920x1080 resolution.

ARRIRAW Test Images

^{ARRIRAW at 200 ISO} ^{ARRIRAW at 400 ISO}

^{ARRIRAW at 800 ISO} ^{ARRIRAW at 1600 ISO}

^{ARRIRAW at 3200 ISO}

ARRIRAW Standard Deviation Calculations
18% Grey Card — 200 × 200 pixel sample

ISO RGB Red
Channel Green
Channel Blue
Channel Luma
Channel

200 2.45 2.24 1.79 2.04 1.80

400 1.89 1.86 1.57 1.66 1.53

800 3.10 2.54 1.84 2.00 1.80

1600 4.18 3.46 2.14 2.25 2.10

3200 4.05 4.77 2.75 3.31 2.81

ARRIRAW Standard Deviation Calculations Graph

ARRIRAW Neat Video Noise Profiles

200 ISO 400 ISO 800 ISO 1600 ISO 3200 ISO

Many thanks to Dani Sanchez-Lopez and Toni Abad.