Canon develops 4K image sensor with 148 dB dynamic range (24 f-stops) - but not for filmmakers
[16:14 Thu,26.January 2023 by blip]
A newly developed HDR image sensor from Canon is currently being talked about by various blogs as an upcoming miracle sensor that will catapult the dynamic range of still and cine cameras to unimagined heights. While Canon's touted "industry-leading" 148dB - the equivalent of about 24 stops - sounds enormous and the technology behind it is very interesting, it should not be overlooked that this 1" CMOS 4K sensor is explicitly designed for surveillance purposes - with not a word Canon itself hints that the new approach should or can also be used for imaging. This is not to say that this is unthinkable, however, the current sensor is probably not very suitable for (scenic) filming from our point of view.
New 4K image sensor with 148dB dynamic range from Canon
The sensor owes its remarkable dynamic range capabilities to a new readout method in which the sensor field is divided into 736 tiles, which are exposed individually - the brighter a subarea, the shorter the exposure time. The sensor is equipped with several processors to be able to determine the exposure times of the many partial areas simultaneously and sufficiently quickly.
Readout method
Unlike conventional HDR processes, this does not require several entire, slightly time-delayed exposures with different exposures to be combined into one resulting image, but the "tiled" image is available immediately, which favors higher frame rates and resolutions (the current sensor resolves at 12.6 MP). Even in extreme exposure situations, this method makes it possible to clearly image both bright and dark areas of the image at the same time, without over/under exposure or motion blur. The supported difference in brightness ranges from 0.1 lux to approximately 2,700,000 lux.
For example, faces can be imaged, tracked and recognized in sharp focus throughout. Even if a car drives out of an underground parking garage, for example, both the license plate (in bright sunlight) and the driver (in the shade) remain clearly recognizable.
This is certainly a dream come true for surveillance cameras, but in scenic filming, it's not just the dynamics that count. We can hardly imagine, for example, that an image consisting of over 700 differently exposed tiles will have good color consistency. At least not before it is subsequently analyzed and corrected by Deap Learning-supported computational imaging algorithms. So we are unlikely to see this new sensor technology in the next Cinema EOS camera. A corresponding approach would be more conceivable in a future smartphone, where manufacturers are always trying to compensate for the disadvantages of a similarly small sensor surface.