Previous news story    Next news story

Exclusive: Fujifilm's phase detection system explained

By Richard Butler on Aug 5, 2010 at 16:31 GMT

With the announcement of the Fujifilm F300 EXR and Z800 EXR coming on a day that also saw four other cameras being launched, it would be easy to overlook their most radical feature. Because, with the latest version of its EXR sensor, Fujifilm has achieved something that's been hoped for but not previously brought to market - through-the-lens phase detection autofocus on a compact camera. The company is claiming the system enables focus times as fast as 0.158 sec. We got some more details of the system from Hitoshi Yamashita, manager of the company's Technical Support Group.

What's so special about Phase Detection AF?

All current compact and mirrorless cameras use contrast detection AF, where the lens is racked back and forwards until the camera finds the position that gives the greatest contrast (which signifies being in focus). Their lenses tend to be designed with very light focusing elements that are fast to move so that this process of trial-and-error can be conducted as quickly as possible. Intelligent processing also attempts to minimize the need to hunt too far for focus but there is still some hunting to be done.

Phase detection, the primary focus method used in most DSLRs tends to be faster but works very differently. To understand why what Fujifilm is trying to do is so desirable, it's first necessary to understand how it works.

How does Phase Detection work?

The easiest way of understanding phase detection is to think about what would happen if you only considered the light entering the far left and far right sides of a lens as it focuses an image.

Back Focus
In Focus
Front Focus
In this simplified schematic, you can see what happens to the image cast by the light passing through the left (blue dotted line) and right (red dotted lines) sides of the lens.

When in focus, the light from both sides of the lens converges to create a focused image. However, when not in focus, the images projected by two sides of the lens do not overlap (they are out of phase with one another).

Of course this is a massively simplified diagram with a single, vertical straight line as the subject (and no inversion of the image as it passes through the lens). The point is that we can derive information about focus if we can separately view light coming from opposite sides of the lens.
 
How does a phase detection sensor 'see'?

And we don't need the whole image to do this. Think about a strip of pixels taken from the sensor in the previous diagram. If you could make one such strip that receives light only from the left hand side of the lens and another that 'looks' only to the right-hand side of the lens, then you have enough information to find focus.

By comparing images from just these two strips it's possible to work out not only how far but also in which direction the lens needs to be moved to bring them into phase.
Back Focus
In Focus
Front Focus

As you can see, the key thing required for phase detection is the ability to capture light only from certain parts of the lens so that (at least) two distinct images can be formed and compared. This information makes it possible to calculate exactly where the lens needs to be moved to, without any of the trial-and-error of contrast detection AF.

It also makes continuous AF easier, since the system can quickly calculate a new focusing distance (and even assess the subject's rate of movement), whereas the contrast detect system has to hunt again, during which time the subject is still moving, meaning the point of highest contrast will keep changing.

Traditionally, phase detection in DSLRs has been conducted by a dedicated AF sensor. The main mirror on DSLRs is slightly translucent which allows some light through to a second mirror and down onto this sensor. This light is split by a series of prisms and lenses in front of the sensor so that different lines of pixels on the sensor receive light from different parts of the lens.

As soon as you see this, you can understand why conventional compact cameras (and all the current mirrorless cameras) have to use contrast detection autofocus. Their sensor elements (sensels) are designed specifically to catch light from all directions and all areas of the lens at once. As such they have no way of separating the incoming light into distinct images. Even with the fastest, lightest lens elements, the trial-and-error nature of contrast detection tends to make it slower.

How does Fujifilm's system work?

Fujifilm's system provides a seemingly simple solution - masking-off half of a sensel means it only receives light from one side of the lens. By creating strips of these sensels, half 'looking' one way, half 'looking' the other, the camera gets the two distinct images necessary for phase detection AF.

'The structure is equivalent to the two light paths divided by two dedicated lenses in conventional phase detection AF device in DSLR's,' says Mr Yamashita.

Although he wouldn't be drawn on the exact arrangement he does give some detail: 'The AF sensels are only arranged in the center area of a CCD, so when phase detection AF is activated the AF point is fixed to the center of an image.'

Of course this means some sensels are receiving 50% less light than their neighbours. Yamashita suggests it need not be a big problem: 'we use several tens of thousands of pixels in the center area of a CCD, which is a very small number of pixels compared to the 12 megapixels used for imaging.' And, he says, they don't simply go to waste when taking pictures: 'sometimes they are used to compose image data and sometimes not, depending on the situation.'

In low light where there isn't enough light to allow phase detection (since the focus sensels are only receiving half the available light) or when face detection is needed, requiring the camera to focus away from the center point, the camera switches back to contrast detection AF. As such the company is billing the system as 'Hybrid AF,' but that shouldn't disguise what is a potentially significant development for compact cameras. We'll be interested to see how the promise of this system works when we get a chance to subject it to testing.

Comments

Total comments: 4
Team Thor Expeditions
By Team Thor Expeditions (Jan 12, 2013)

Great suggestion, Doug. Perhaps the actual shipping system could use two small lenses as you described just for focusing (similar to a rangefinder system where the distance between the centroids of the lenses is "magic") and the third lens for the actual imaging. This might provide the best of all worlds.

0 upvotes
Doug Bale
By Doug Bale (Jan 11, 2013)

As this article's final paragraph says, phase-detection becomes less possible as light levels lessen. It also grows less practicable as cameras get smaller, with less distance between opposite sides of their smaller lenses. Fuji's technology is good news for cameras the size of a half-brick, less so for the subcompacts that are the main challenge to DSLR dominance.

But tiny lenses have their own advantage: they can be made in excellent quality at lower cost (as in smartphones). Why has no camera-maker yet returned to the twin-lens rangefinder paradigm, putting a small second lens 60-odd millimetres to one side of the main one and electronically superimposing part of the image from it onto the central area of that from the prime? — no more difficult to do than superimposing the control display over a Youtube video. Electronics monitoring the view covered by the prime could determine what portion of the secondary's image should be used in the superimposition. Let's hear it for Euclid!

1 upvote
solobo
By solobo (Jan 18, 2013)

I like your suggestion, except wouldn't that require 2 sensors as well? Not so cheap...

0 upvotes
Grzegorz Popiela
By Grzegorz Popiela (Mar 3, 2013)

So does that mean that two-lens 3D cameras are capable of phase-detection auto focus by default? Or let me rephrase - so instead of making a bulky SLR camera, if we add a small lens to the side of MILC, a pop-up lens like a pop-up flash, we can enable faster AF and get 3D photos as a side effect? Nice! :)

0 upvotes
Total comments: 4