I was having a brief rant in a previous post about the new iPhone 4’s supposed HD video capabilities. Now there’s a few in the wild I managed to procure one long enough to wave it about at some vertical building to see how severe the rolling shutter artefacts are.
It comes as no surprise to discover that it’s not that great – actually it’s fairly dire – have a look at this frame grab of a van going past above.
So what’s wrong with that you may ask? Well, the orange diagonal line should really be vertical if the image had been captured correctly. Below is the footage posted on Vimeo so you get a better ideas of what I’m talking about:
Right, so what exactly is going on here?
In a proper video camera each whole frame is captured at one instance in time – all of the data is read out from the sensor at the same moment – usually a 25th of a second. This gives a perfect image but is quite bandwidth hungry – there’s a lot of data to be moved about for each complete frame to be captured.
Video implementation on most cheap phone cameras by contrast “cheat.” By processing each line from the sensor sequentially over a period of time you can cut down on the bandwidth requirements. The problem with this is if anything is moving though the shot vertical lines appear to shear (as above) because each line within the the picture is captured a fraction of a second later from the top of the image to the bottom.
So is this a complete disaster? Well, the answer rather depends on what you intend to do with the footage and the type of material you intend to shoot. Ultimately, the iPhone is a domestic product designed to be used for capturing home movies – of course that’s not going to stop people trying to use it for purposes other than that…
Interestingly there is a product designed to fix this issue – Rolling Shutter from The Foundry, although as it weighs in at several hundred bucks I suspect that we won’t all be rushing out to buy it.
UPDATE: There’s a free trial of the plugin from The Foundry, so I stuck my footage though it to see what the results would be:
I think it’s fair to say that this is an improvement but at the expense of some really nasty artefacts. However, this is a really vicious test for any piece of software to try and cope with and I think it fairs pretty well.
On the down side are the rendering times – I clocked this clip at about 15 minutes on a bog standard desktop PC. It’s not something you’d want to be doing too often then.