Posted by: recordingsofnature | April 14, 2016

Variable Anti-Moire and Anti-aliasing Filter for DSLR digital video – DIY

This is an update on the CYL-filter method for anti-aliasing of DSLR video, and a description of how to get a filter which has a variable anti-aliasing filter strength. The method is to simply use two filters in a stack, and by adjusting the rotation it is possible to vary the filter effect from zero to the double amount. Almost like a variable ND filter.

In this article I will sum up the developments, and give a demonstration of a prototype filter. Unfortunately, I am not able to do any manufacturing or selling. Instead, I will try to give all details to make way for trying it DIY.

Variable anti-aliasing filter (CYL-filter) for digital DSLR video

The original CYL-filter method has limitations in terms of a strict working range with a narrow set of working points with respect to aperture and lens focal length. A very strong filter is normally needed for wide angle lenses, while a very weak filter is needed for long lenses. Add to this the dependency of the aperture size. So, a variable solution is really an advantage and extends the working range.

Background

The job of the anti-aliasing filter is to filter out high frequency content of the image (low-pass filter) so the information matches the pixel resolution of the image sensor. This is needed for an optimal sampling according to the fundamental sampling theorems. A suboptimal sampling process (a common problem especially with color-masked and line skipped DSLR image sensors, where the built-in OPLF is of less use) will cause irreversible image degradation and artefacts affecting throughout the subsequent digital image processing chain. Aliasing noise and artefacts are particularly problematic in post-sharpness algorithms, video compression and color control.

In the race towards higher pixel counts in digital video, spatial aliasing (pixel fill factor, super sampling) and temporal aliasing (shutter angle and fps) are often less prioritised. Aliasing issues are highly responsible for the pronounced “digital” appearance of many digital videos, with exaggerated sharpness, flickering non-smooth motions, various digital noise and Moiré patterns. Moiré patterns are the most visible cues of aliasing, appearing in scenes with structured and repeated high frequency content, such as for fabrics, brick walls, wire grids, etc. In other scenes the aliasing can be less noticeable at first sight, but a closer look will reveal that the image suffers from strange digital noise especially in scenery with a lot of high freq content,  such as fur, grass, tree canopy and detailed landscapes etc. to mention some of the worst challenges.

Demo prototype

The CYL-filter works by introducing a tiny amount of optical blur through the use of cylinder (astigmatic) lens elements in front of the camera lens. The cylinder axis of this lens must be at a 45 degrees angle to the sensor axis. The double filter is composed of two equal filters (this time CYL75 strength) clicked together. That means, it can readily  be detached and used as single fixed filter as well.

The video demonstrates a prototype of 2 x CYL-75, on a 77mm frame size Sigma 10-20mm lens on Nikon d5100. More specifically, the testing is at f/5 for f= 13mm at at 85% filter, and 18mm at 60% filter.

For best result, watch the video in full-screen and at highest quality.

Evaluation

Id like to give a critical and detailed review of the performance, as I know this method has both pros and cons.

  • I think optical performance of the filter is close to, and in line with the performance of the single CYL filter. Varying the strength gives an interesting freedom to explore the optimal sharpness depending on the scene. It’s also really an advantage to have an extended the range of e.g 2 x 75 = 150 which is needed to operate in for wide angle lenses.
  • At filter strength 0% the image sharpness and image quality  seems is nearly unaffected which is a good sign. It is likely that some finer lens tests may reveal some image degradation.
  • The filter produces a bit of 45 degree barrel distortion which is proportional to the filter strength. At 0% filter strength I dont see any distortion. It starts to get noticeable at maybe 50-70%. I have noticed that this distortion can be reduce by placing the filter as close to the front lens element as possible.
  • Edge sharpness may seems a bit soft for the very wide angle tests.
  • Adjusting the strength requires each filter element to be counter rotated, and it is currently a little clumsy, as the marks on the filter rim are not so clear. The bulkiness of the 77mm double will also prevent a ND filter to be mounted on top.
  • As the filter uses 2 lenses  (ophthalmic, multicoated CR32 lenses) there could easily be some image degradation, which could be critical in some applications. It is very important to keep the lens surfaces clean. A lens hood is a good idea to reduce possible additional lens flares in sunny conditions.

To improve the filter, it will be crucial to use high quality, dedicated optics instead of currently  ophthalmic lenses (high quality multi coated CR32). Using optimized optics, with optimal curvature and thin shape will also allow a more compact filter design. This should also reduce the barrel distortion and improve edge sharpness.

I think the real optimal solution would be, if a lens producer could implement this filter method inside the camera lens, by adding the cylinder lenses to two existing (rotating) lens elements inside the lens. It could be an additional focus option to control the anti-aliasing effects of the lens. This may also effectively eliminate the barrel distortion, however, it will need some optical simulations to determine that.

DIY details

I’m really not able to produce this filter for selling. I cannot currently produce the product in any proper, reliable quality and my 3D printed frames are not really sturdy or durable. Instead I will just give details for you to try it DIY.

Lens frames and mounting

The lens frames are 3D printed from this design which fits on my 77 mm diameter Sigma lens and uses an uncut 70 mm filter lens. The design can be downloaded here  (zip file containing STL, FreeCad, Step formats. If needed, contact me for SW format)

Design of frame for varialbe CYL filter, antialiasing filter for DSLR digital video

In the design, r1, r2 and r3 are the main parameter to control size of the frame. These and other dimensions will probably need to be adjusted depending on the performance of the 3D printer. I was able to print out this design (stl file) on a Wanhao Duplicator 4 directly and without the use of support structures (ReplicatorG settings: PLA, T(plate)=55C, T(nozzle)=215C, feedrate=50, travel feedrate =50, layer 0.27mm, fill 15%). But in general, it is always cumbersome and may take several attempts to achieve correct dimensions and good fitting of the lens to the 3D printed components. This all depends on how the 3D printers handles the sub-millimetre details. A good strategy is to have it slightly too tight and polish off material until the click-ons and press fits are good. The optimal solutions is when the lens just snaps in firmly, without the need of glue or similar. If too lose, may try to add some nail lacquer on the lens edge. It should not harm the lens. Use of most solvent based glue on the 3d printed PLA will easily deform the structure. Superglue is normally very strong, but can also easily break the lenses.

Optics

It should be possible to buy the lenses from your local ophthalmic at a good price.

Eg. Hoya HILUX, Thin.  Antireflex and Multi-coated. Standard, uncut sizes are 65mm and 70mm.

The lens prescriptions for the CYL filters are:

  • CYL25 :    S-0.25  C+0.25 (S+0.00 C-0.25)
  • CYL50 :    S-0.25  C+0.50 (S+0.25 C-0.50)
  • CYL75 :    S-0.50  C+.75   (S+0.25 C-0.75)
  • CYL100:    S-0.50  C+1.00 (S+0.50 C-1.00)

The lens size will need to be cut according to the frame size used on the camera lens. For the present 77mm frame size design, an uncut 70 mm lens can be used with no modifications. You may ask your provider to cut out the lens in a special size if you will avoid to cut it yourself. Possible by providing him with the empty frame. It is possible to cut the lens size yourself by use of a fretsaw (with good support and a firm hand) and a gentle rotating grinding machine. The lenses are fairly durable and sustain touching and can be cleaned with hand dish washing detergent and ethanol.

Alignment

Precise alignment of the optical cylinder axis at 45 degrees to the lens frame, as shown on the drawing, is important. You can ask your lens provider to accurately indicate the cylinder-axis directions on the edge of the lens. Or you can determine it yourself by use of simple setup with a laser pointer and a binocular, see diagram below.

Method for determining the axis direction of cylinder (astigmatic) lens

When you know the direction, mark it with an small indent in the lens edge to be able to see it clearly.

Use

With the double filter each of the two lenses need to be counter rotated by the same amount, in order to maintain the same resulting cylinder axis, at 45 degrees. If the lenses are equally aligned the strength will be double, and when counter aligned by 90 degrees the effect will be eliminated and the resulting filter strength approaching zero. In between there is a gradual variation.

2 filters rotation_v2

It is difficult to give a precise formula for the filter strength needed for different lens settings and cameras. The strength depends on aperture and focal length (zoom), and different cameras will need different amount of blur to remove the aliasing. It is best to build up your own table of good settings where the filter is just right, based on your own experience. It could look something like this incomplete table below (these values may not be accurate for other camera/lens setups).

filter table

Advertisements

Responses

  1. […] – See this article for update with latest […]

    Like

  2. […] [Update April 2016 – DIY variable antialiasing filter method – here] […]

    Like

  3. […] [Update 2016 April, DIY variable anti-Moré filter method for DSLR] […]

    Like

  4. Dude, would love to buy this to try on canon 5d mark iii when using magic lantern raw dual iso feature (each horizontal line is different iso, which gives you up to 14+ stops of DR, but creates terrible aliasing).

    If I can afford it, I’d be glad review on the forums for the thousands of ml fanboys that want to use dual iso but have to throw half the footage out from aliasing. This + chroma smoothing in post would be killer if this doesn’t knock it out completely.

    Like

    • Hi Brookes, thanks for the comment.
      Ok, definitely sounds like an interesting approach with the magic lantern raw Dual ISO! I didnt know this feature.

      Unfortunately I have not been producing the filters for some time.
      so I am a bit far away from being able to supply any…

      BR
      Kristian

      Like


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

Categories

%d bloggers like this: