Hands-on with the K-Lens, a light field lens that captures 3D data for post-capture refocusing

We’ve seen light field cameras before, such as the Lytro camera, but K-Lens is the first light field interchangeable lens to be available for existing digital still and video cameras. The lens is able to record a grid of images that contain information about the relative distance of different objects in the scene. This information can then be used in post-processing to adjust the focus point in the scene, as well as to easily mask different areas to alter the brightness and colors of different layers in the scene. The images can also be used to create moderate 3D effects that allow us to ‘see around’ objects as the pictures in the 3×3 grid are all recorded from a slightly different perspective.

Each image taken with the K-Lens is accompanied by a depth map that shows the relative distances from the camera of different elements in the scene. Lighter tones represent closer objects, and darker further

Applications for this kind of technology include focus adjustment in video, the creation of 3D content, dissection of the scene for animation as well as the ability to create shallow depth-of-field effects and easier distance-based editing—such as masking a background to edit it separate from the rest of the image.

What it is

The K-Lens is constructed in three distinct sections that can be seen clearly from the outside. The fromt part of the lens is of a regular 80mm optical construction, but when the light passes through it enters the second section of the lens—the mirror tunnel, which sits were the sensor in a normal array would be. Instead of hitting the sensor though, the focused light meets a series of mirrors in the tunnel that are positioned at different angles to allow each a slightly different view of the subject. The mirrors, along with the light that passes directly through the tunnel, create nine individual images that go on to pass through the rear section of the lens to the camera’s sensor. The result is a grid of nine pictures in a variety of orientations, all recorded on a single frame. The effect is akin to looking through a kaleidoscope—hence the name ‘K-Lens’.

A diagram that shows the path of light as it passes from the front of the lens through the mirror tunnel

The key part of the lens is the mirror tunnel, while the forward end governs the focal length of the system and the rear end how the size of the image matches the sensor size in use. I’m told that changing the focal length is simply a question of replacing the forward section, and that, at least in theory, it’s possible to make this a zoom.

The lens offers a thread on the bottom to allow it to be mounted directly on a tripod. Between the lens and the camera you can see a Nikon F to Nikon Z mount adapter so I could use the lens on the Nikon Z7

The K-Lens idea started out as a mirror tunnel adapter that could be mounted to the rear of a regular lens to fit between the main lens and the camera, and which would make any lens into a light field system. The main problem with that idea though was dust getting into the mirror tunnel caused dramatic issues with lost data. As each image is only a 9th of the full area of the sensor even a small piece of dust can occupy a significant part of the image area. A further issue was that to deliver high-quality depth data a lens needs to be sharp right across the frame as the system relies heavily on light gathered at the edges of the imaging circle, where quality very often drops off.

The optics used in the K-Lens have to be of exceptional quality and the image needs to be much sharper corner-to-corner than regular lenses would produce—which goes some way to explain why it’s so big. The company tried to use a smaller lens but found the required quality could not be achieved. The system can be made smaller when used for a smaller sensor, and the company is currently working on a version for 1in sensors in industrial applications.

Looking through the front of the lens you can see how the image is divided up. Here the rear opening is reflected multiple times in the mirror tunnel

Most light field cameras use an array of micro lenses to divide the image into the constituent parts but with this mirror tunnel design the advantage is that it isn’t tied to a capture device – so it can be incorporated directly into a lens that can be used on regular cameras. The team did also look at using an array of lenses inside the main lens instead of the mirror tunnel, but that would have required an entirely new concept when it came to designing the main lens. Using the mirror tunnel allows existing lens designs to be used, so the company didn’t have to start the lens design from scratch.
The optics in the K-Lens are designed by Carl Zeiss Jena GmbH in Germany, who will be building the whole lens once production begins.

What does it produce?

The kaleidoscopic view presented by the K-Lens records a grid of nine images in different orientations and from slightly different perspectives, all recorded on a single frame

Looking through the viewfinder when this lens is attached to your camera, you see a grid of nine images in different orientations—which can be a bit of a surprise. The recorded picture will look this way too, on the back of the camera and in normal software. When the image is fed into the company’s SeeDeep software application, it gets analyzed and the nine images are used to create a depth map that charts the different distances represented in the scene using a range of tonal values. Once this is done, the ‘normal’ image and its depth map can be loaded into Photoshop, or any program that allows a depth channel, and we can use the depth map to individually edit objects at different depths in the scene.

A depth map produced of a tree. Its leaves are behind it so are represented with a darker tone. The wall in the background is darker again to demonstrate it is further away

The SeeDeep application also combines the information from the nine images to create what the company calls a Super Resolution image which is half the resolution of the total sensor output. Without this option the final image would be only 1/9th of the resolution of the sensor, which obviously would make the whole exercise much less worthwhile.

It’s also interesting to note that the software can still create a depth map even when the nine individual frames aren’t all whole. This is crucial as most cameras will record video from the 16:9 crop of the sensor. The SeeDeep software can still generate enough depth information using the upper part of the lower images and the lower parts of the upper images. Some cameras, like the Lumix S1H, can provide video from the whole of the frame which obviously provides more data and a more accurate depth map. Also of note is that when used with the S1H in 5.9K mode the nine individual images created will have HD resolution.

What can it be used for?

No one is going to buy this K-Lens and use it as a regular lens, so customers will be those who can make the most of the depth information it creates. The most obvious uses are in the area of depth-based editing where the user wants to adjust the focus position or add blur to objects at certain distances to create a shallow depth of field effect.

The Lens Blur feature in Photoshop can pick up the depth map when it is loaded as a channel. The feature then allows us to click on the object we want in focus and allows us to blur the rest of the scene. On the left is the original image, in the center the focus is shifted to the background and on the right the focus is on the foreground

When adding blur using the depth map the degree of blur will be proportionate to the tonal value of that area of the depth map, so more distant objects will automatically be more blurred than closer ones in the same way real optical blur appears. The depth map can be adjusted in software to exaggerate or reduce the ‘depth’ and objects can be added to, or eliminated from, any distance by altering their tonal value in the depth map.

Here I used the depth map to identify the area behind the subject so I could warm it up without altering the coloration in the subject

The depth information can also be used to isolate specific distances for brightness or color editing, so a subject can be warmed and anything behind made cool – for example. Of course this can be done with normal images in software when we create selections, but the idea here is that the layer masks are automatically formed by the depth data in the depth map. This comes into its own in video editing, as creating masks to follow moving subjects and backgrounds is much more difficult. In video the ability to make focus adjustments is also extremely useful and can save shooting a scene again if the focus was missed the first time. Focus pulls can also be achieved in post-production, which would be impossible to do without the depth information the lens provides.

The color changes from frame to frame in the video above will not feature in the final product.

The different perspectives recorded by the lens can also be used in the creation of 3D content where moderate differences in a perspective are enough for the required effect. As the virtual baseline for 3D is only 1.5cm the lens works best in 3D with very close subjects and for macro photography as the baseline will be able to ‘see around’ objects better than when subjects are further away. The closest focus distance of the lens is 80cm, but it can also be used with macro extender tubes. In general though, depth data is most accurate and reliable in subjects closer than 10m. The degree of 3D is also just about enough to create ‘3D’ images on Facebook of the type some smartphones can also produce from multiple lens arrangements.

The virtual baseline (the perspective shift between images recorded at the edges of the nine-image grid) depends on the aperture of the lens, the mirror dimensions and the magnification of the pick-up lens. So it will be different for different focal lengths. The effect will be more exaggerated in longer lenses, so long as the physical size of the iris opening remains the same.

The company says devices such as the Lume Pad and Looking Glass displays also work well with images recorded using the K-Lens, and 3D effects can be viewed using Google’s Cardboard. Relief printing, where the image is printed in 3D, is also an application, as is lenticular printing.

In use

The squeezed area of the barrel makes it obvious there is something different about this lens, and this is where the mirror tunnel is housed. The lens shown here is a very pre-production version, and the aperture scale is reversed. The longer line though indicates where the company recommends we set the iris – in the 75% closed position

The lens is a lot bigger and heavier than a regular 80mm of course and it is best used on a tripod – though it isn’t so awkward it can’t be handheld. While the size and weight are good reasons for tripod use on their own, another is the small apertures required to get the most from the lens. There is a scale on the iris but it isn’t marked with f/numbers as the company says traditional aperture markings aren’t relevant. There is a larger marker where the company says the lens works at its best, and this is with the aperture closed to 75%. This delivers an extensive depth of field and enables the focus to be shifted more easily in software afterwards and provides the best depth data.

We are told that the aperture is just a means to let in light, not to control depth of field or out-of-focus highlights, and it is better to increase the ISO of the camera than to open the aperture more. There is no information about maximum or minimum aperture settings, but I do know the iris uses 15 blades for a really rounded aperture.

An optional extra, the K-Lens monitor automatically presents only one of the nine images created by the lens. This makes composition and focusing much easier, though it also adds to the bulk of the system

Looking through the viewfinder we are presented with a kaleidoscopic view of the scene in front of the lens. This makes focusing and composition quite hard, so we need to magnify the view until one of the nine images fills the finder – the image at the top left is in the correct orientation. An alternative is to use the optional HDMI monitor provided by K-Lens which displays this part of the image by default.

Using the depth map as a layer mask allows quick and easy editing of areas at different distances from the camera. Here, for demonstration, I’ve used the mask to make the background darker

When the images are fed into the company’s SeeDeep software we get a regular colour image and a depth map returned. I used the Nikon Z7 which has a sensor resolution of 45MP. Each of the nine images is 5MP, but the software can create Super Resolution images that are about half the resolution of the sensor, so in this case the Super Resolution image and the depth map are 17.6MP. The Super Resolution images look very good, as they aren’t created by standard interpolation but by using all the information in the nine smaller images to fill in the details that need to be added to increase the resolution. The same Super Resolution techniques can be applied to video footage, allowing SeeDeep to produce resolutions of up to 4K.

Conclusion

The lens hood is pretty unusual, with a heavy lip on the front end. This one is pre-production and 3D printed but the production design will stay the same. I couldn’t get a good explanation for its shape, but was told the design was specified by Carl Zeiss Jena as part of the optical design.

This is clearly a very specialist lens that will be useful in a more limited range of applications than regular lenses, but which really does offer something new. The world got very excited about the Lytro cameras but the idea was ultimately undone not by the shortcomings of the technology but by reliance on a dedicated cameras and the low resolution of the images it produced. This concept allows users to employ the cameras they already have and all the features they offer, as well as much more practical working resolutions. The fact that any full frame camera can be used opens the market to a much wider audience, increases the incidence in which it can be used and reduces the amount of learning required.

The lens I used was very pre-production and I couldn’t get to use SeeDeep myself, but even in with these limitations I was able to get an idea of the effects that can be created, and the relative ease with which they can be achieved. The K-Lens certainly has promise and I look forward to seeing what photographers in the field can create with it. Once this one is off the ground it will be interesting to see what other focal lengths the company will offer, and how the technology will develop. I have been told that K-Lens hopes to be able to integrate ND filters into the mirror tunnel to enable one-shot HDR imaging by recording a wide range of exposures in a single shot – which would be another useful feature.

This lens isn’t going to be mass-produced, so buyers can request the mount they want. Initial samples are in Nikon F and Canon EF mounts, but K-lens says most other mounts will be easy to accommodate

The lens will retail for around €3599/$4199, but can be had at a 45% discount with an ‘early bird’ offer. The optional monitor will cost €299/$339, but again is €199/$229 at the moment. Production is due to begin in January 2022, and shipping will start in August 2022. For more information see the K-Lens website and the company’s Kickstarter campaign page.

Press release:

Note/disclaimer: Remember to do your research with any crowdfunding project before backing it. Pledges to crowdfunding campaigns are not pre-orders. DPReview does not have a relationship with this, or any such campaign, and we publicize only projects that appear legitimate, and which we consider will be of genuine interest to our readers. You can read more about the safeguards Kickstarter has in place on its ‘Trust & Safety’ page.