CMOS sensors and rolling shutters

[Mark Donnell]

I am unclear about the association between these two - do all cameras using CMOS sensors have rolling shutters, or is it just the lower end group that puts these two together ?

[Mark Donnell]

I have done further reading, and would refer others interested in this topic to the Wikipedia entry on "Rolling shutters". It has great links at the bottom of the article, including one to a B & H Photo article, which references Barry Green's thorough review of this topic. In short, all current CMOS sensors do use rolling shutters. These articles also outline the types of distortion common to CCD and CMOS sensors and their associated shutters. It was very interesting reading.

[Jack Zhang]

I'm an advocate for Global Shutters on CMOS sensors. Omnivision (the people that make the iPhone's back camera sensor) claim to have a patent on a mixed Rolling/Global shutter design that has yet to be made. (Nope, 4S doesn't have it.)

Meanwhile, Viimagic has a 2/3'' global shutter CMOS sensor for sale, but only Chinese markets are interested at the moment since that's the only place they've shown it off at currently.

The problem with Global Shutters on CMOS sensors is noise and cost. It used to be that because the diodes had to sit below layers of active matrix/wiring, there was high amounts of noise because of little light hitting the diodes. This is an example of such noise on prototype sensors: Low-Noise Global Shutter Pixels for CMOS Image Sensors (IDL)

Increased build quality inherently solves this, at the cost that it be impossible to reach for consumer manufacturers.

However, progress is being made to drive costs down. The BSI sensor (or back-illuminated sensor) allows more light to reach the diodes. Omnivision's patent is showing they're trying to combine the technologies, however, (even in the patent) it is a switchable system compared to a fully Global Shutter system.

We'll still have to wait and see what comes of this at NAB. At the moment, everyone's coping with it and the problem of flash banding, skew, and etc is here to stay for the next while.

[Pavel Sedlak]

The 2011 Zacuto shootout Part 3 is finally released! Watch and enjoy! | Philip Bloom

There is a part about rolling shutter (5DMII, 7D, F3, AF100, ...)

[Mark Donnell]

Thanks for your comments, Jack. I will hope that global shutters will come soon to CMOS sensors. Pavel, thanks so much for the link. I was unfamiliar with this series, and it was fascinating ! For me, the bottom line is that, in sensitive tests, rolling shutters and 4:2:0 color spaces are clearly inferior to global shutters and 4:2:2 color spaces.

[Jack Zhang]

I was surprised by the Weisscam and Phantom Flex in the shootout. They are rolling shutters, but they're so fast, it's virtually a global shutter. I would only expect artifacts if you shoot above 1000FPS with these cameras.

I also completely agree with Solomon Rogers when it comes to Rolling Shutters and matchmoving. It literally consumes more time in post and it's a gigantic headache to deal with. (Around 08:53 into the video)

Again, this is why you don't really see any rapid camera movement in the early days of shows like Sanctuary, because they were using the RED ONE, which has these rolling shutter problems.

[Chris Medico]

I wish CMOS with global shutter was the norm. It was good to see in pt3 of the shootout that the artifacts could be minimized.

On a side note - Jack, I feel the same way about automatic transmissions in cars as you do about rolling shutters. ;)

[Mark Donnell]

It seems clear that some rolling shutters have such quick readout speeds that only in cases of extreme camera or subject motion will the rolling shutter become a problem. We really need some enterprising individuals to quantify this for some of the newer cameras that are now on the market. For instance, if one is shooting with a stationary camera at a shutter of 1/100th and a subject motion of 30 MPH transversely across the frame (like a car), will any rolling shutter artifacts be of concern ? This could be predicted mathematically if we knew the sensor read-out speeds, but to my knowledge, read-out speeds are not available.

[Jack Zhang]

Flash Banding will always be a concern. I tried FBC on the HPX300 and it only does a decent job, It doesn't patch up all the flash banding in real time.

[David Heath]

Originally Posted by Jack Zhang

I tried FBC on the HPX300 and it only does a decent job, It doesn't patch up all the flash banding in real time.

Worse, it can give false corrections - not many, but enough to be a problem. And a false correction can look FAR worse than an uncorrected rolling shutter artifact.

Hence far better to do any such corrections in software in post - if it looks better, fine - it it really screws up, reject it. That's not an option with the real time method that Panasonic use.

[Cole McDonald]

I was just thinking one could build a rig that differentially dutched the camera slightly as one panned to counteract the rolling shutter effect.