Red rolling shutter redux
There’s been a bit of discussion about the image distortion caused by the scanning (rolling) shutter used by CMOS chips which are starting to proliferate in the prosumer and professional video camera world. The key problem, as I have mentioned previously, is that the scanning imaging device is no longer sending the image solely to scanning displays- i.e. tube televisions. Today’s displays include plasma, LCD, DLP, OLED, etc. Most are progressive, but some include circuitry to display the image as if it were a scanning device.
Mike Curtis, of HD For Indies, is very heavy into RED usage and promotion on his web site. RED is the video camera that I will agree is changing, or will change, the hardware business in the video industry. I checked in with Mike about the “rolling shutter” issue…
He says that the CMOS chip in the RED camera “scans- but quickly.”
Since the RED camera is designed to emulate a film camera, with an adjustable angle gate that effectively allows shorter or longer exposures of light onto the film/chip, and the camera is capable of shutter speeds from 24 to 60 fps, I wondered what the actual scanning frequency of the chips is. Because it could scan slowly, quickly, repetitively (for slow shutter effects) but I’m just guessing.
When pressed for information, he noted that it was “most noticeable in extremely fast events – we saw it in flash photography” which, I will admit, is an extremely fast event. Even with CCD camcorders, a 1/50,000 of a second strobe flash only illuminates one field of video (sometimes causing problems). The flash does not last anywhere near the 1/60 of a second a single field takes to be captured. So flash photography is a special situation. Moreover, there’s already discussion online about how to minimize flash issues by timing the flash with the camera shutter- something you have to do with film too.
While users go pouring over various images and clips looking for errors, be aware that not all CMOS chips (and the controlling systems) are created equal. There has been some very good discussion in the RedUser.net forums about this. Various frame grabs (visible here) show whip-pans where vertical lines are all vertical. Blurred from camera motion, but not distorted or diagonal- two of the most awful artifacts. Kudos to the RED engineers for minimizing the visibility an inherent CMOS process.
Even so, a global shutter would accept data from all the imaging pixels at the same time, bank it, and then send it off to the codec chips. This would eliminate any distortion (like this, or this, or this, or this) when viewed on a progressive display or when single frames are exported.
Can it be fixed with a firmware upgrade?
The chip is designed a certain way. The active pixels, masked pixels and associated registers are built according to a design that really can’t be rejiggered after the fact. Now, maybe the software can look at the data coming off the chip, know that each line of data is 1/250th of a second behind the previous, and use algorithms to adjust for motion between the lines. But it can’t make up for camera movement, where the image will be falling off the edge of the line. it can’t make up for fast camera tilts, where lines could have redundant data or the image moves so far between the scan lines that each line does not bear any relevance to the previous.
So, for the most part, I’d say no. But that doesn’t stop exceptionally smart engineers from developing software that can do amazing things. Hopefully, we’ll see CMOS with global shutter (true progressive) sooner rather than later. Till then, you can expect better camcorders to do a better job with rolling shutter issues than cheaper camcorders. You get what you pay for.