HubertChen: At this size and price and battery capacity the most straight forward way to reduce noise is simply to cool down sensor and electronics with a peltier element. Works wonders for your Signal to Noise ratio and is very easy to do. It is simply not done in consumer cameras due to size and battery usage.
And you need a radiator on the camera, or a cooling fan, to transport the heat out of the camera. The radiator could be placed on the camera's underside.
DDWD10: I'm a little concerned about the 1/500s mechanical shutter limit. Using the electronic shutter at higher speeds, would one expect CMOS rolling shutter artifacts or not?
Yes, off course, but mechanical shutters have exactly the same problem. Short exposure times are realized by a narrow slot that moves over the sensor rather slowly.
The important aspect of the shutter is the flash sync time, typically 1/60 s in cheap models, to 1/250 s in professional DSLRs. Any faster exposure times are realized by a slot that moves over the sensor, and lead to the characteristic rolling shutter artifact.
T3: I remember there was a big discussion a while back where people were claiming that the key to lower noise was having more pixels, and of smaller size. Wonder where those people are now?
For each pixel, a certain amount of noise is created, each time an image is taken. This noise originates from imperfections in the analog circuitry and from the analog to digital converter. This noise is called "read noise". It is independent of the amount of light.
As you are writing about the amount of light, I think you are confusing read noise and shot noise.
Shot noise is an other source of noise that comes from the quantum nature of the world. The world itself is not smooth but made of particles that are subjected to random. Photons, for example, arrive at random points in time at a certain pixel. This random arrival of photons creates shot noise, which is significant in low light.
I write averaging because averaging is at the core of most noise reduction and down sampling algorithms. These algorithms usually contain smoothing steps which are usually weighted averages of neighboring pixels.
David Elliott Lewis: This low light sensor is an amazing technology that appears to extend photography into areas previously unavailable - providing high definition video in near total darkness - light so low that even the unaided eye cannot see.
I believe there is huge commercial potential including in motion picture production. I hope Canon will bring this sensor to market.
One suggestion would be to create a dual sensor camera containing both this new low light sensor and a traditional high resolution sensor mounted on a rotating device that can quickly switch between them. Possibly, the same type of mechanical device that moves neutral density filters into place could be used alternate between sensors.
P.S. The reason I am suggesting a dual sensor camera design is based on the assumption that it will not perform well under daylight conditions. With such a low pixel count, visual artifacts like moire might degrade the image.
An other idea for switching sensors, would be a mirror.
This could be a new application for the mirror in SLR cameras, and would leverage the the properties of existing SLR-lenses in the age of the electronic viewfinder.
A possible camera could be constructed like this:
The high resolution sensor is in the top of the camera, light reaches it via the mirror. The high sensitive sensor is in the back of the camera, light reaches it when the mirror is moved upwards.
This kind of camera could even have a separate focus sensor, with a similar arrangement like in current SLRs: The primary mirror is half transparent. There is secondary mirror behind the primary mirror that directs the light downwards to the focus sensor, which is in the bottom of the camera.
These people never knew what they were talking about. Or when they were from the industry, they tried to protect their work on ever smaller pixels.
Big pixels are necessary for high sensitivity - unless you can accurately count photons. But as long as there is read noise each pixel introduces additional noise into the image. You can reduce the read noise by averaging neighboring pixels but this is much less efficient than big pixels.
The only noise that can be efficiently reduced by averaging is shot noise. Shot noise is additional noise from quantum mechanics - the phenomenon that photons are individual particles that arrive at random.
Robert Morris: Aw, the price we pay for all of our gadgets.
The Chinese are paying the price! We get the gadgets cheap, and our air quality is quite high.
I hope this camera will be successful!
All future SLRs should have WLAN and cellphone connectivity, GPS, and the ability to install apps. A serious camera should have separate buttons and wheels for all important functions however, and not a touchscreen interface.
The additional connectivity, and the GPS, would be very useful for journalists.
Apps could be very useful for special tasks like animal photography. An app could release the shutter if something new appears in the frame. The camera could be remote controlled, or it could be remotely checked if there are interesting photographs on the camera.
I would like to have a map app on my camera for hiking. Additionally such a camera could be used as a phone for emergencies. As batteries for SLRs can potentially be fairly big, such a device could run for very long on a single battery.
Caleido: Come on. This is down right ridiculous. The idea, the look, the results. Everything.
No this is a great idea! It is a tripod and a light tent at once. However you could make it yourself out of a large yogurt or ice cream container.