Posted by: recordingsofnature | September 25, 2012

DSLR video Antialiasing filter – Sharpness vs. moiré

This article follows up on the antialiasing filter method described in the earlier post  Antialiasing filter for DSLR video, for reducing or eliminating moiré video artifacts typical for DSLR HD videoing, -quite visible on the Nikon d5100 DSLR camera. Now, I finally managed to do a more through-out test to map the antialiasing effect and sharpness for different filter and lens settings.

[Update 4. dec. 2012 - the filter mounted in a standard filter frame is now available - here]

Avoiding aliasing and moiré artifacts is all about avoiding a too sharp image to be formed on the image sensor.  The antialiasing filters must introduce a precise amount of blur, whereby resolution of the image on the image sensor corresponds to the resolution of the recorded video. This can also be described as a lowpass filter.

Example of the filter effect on moiré patterns

Example of the filter effect on moiré patterns from the video clips at the bottom of this article

The type of filters used here works by introducing small amounts of blur through astigmatism, but since this effect is highly depending on both the aperture and focal length, some sharpness tests were needed to provide an overview of the effects. The following tests are based on still images. To make the results more applicable, the amount of sharpness has been converted into a number corresponding to the maximum vertical pixel resolution for the 16:9 picture crop area.  This mean that values of about 1080 pixels should be a good start for a 1080p HD video. When the value is too high aliasing will occur. Too low, the image will start to blur. However let’s see the results.

The first chart shows the maximum pixel resolution in vertical direction with no filter.

Vertical pixel resolution for 16:9 crop area without filter

Vertical pixel resolution for 16:9 crop area without filter as a function of the focal length (x-axis) and aperture (y-axis)

This is the maximum resolution measured on normal still images using the Nikon 18-55mm f/3.5-5.6G AF-S VR DX  and Nikon AF NIKKOR 28mm f/2.8D. This can be compared to the resolution of stills; 4928 x 3264 (= 16MP DX sensor), and taking the 16:9 crop area gives 0,75 x 3264 = 2448 px in the height. So, this is the actual maximum resolution for this camera system.

It can be noted that the sharpness generally is much higher than 1080, so there is good basis for plenty of moire and aliasing artifacts. I can also be noted, that for very low apertures, f/20-22, and longer focal lengths, f=35-55 mm, the resolution is clearly dropping due to diffraction, thus creating a natural blur effect.

The following charts show how the maximum sharpness varies when the anti-aliasing filter is applied in front of the lens. Filter strengths are 25, 50 and 75. The missing values in the chart relates to the available lenses used in the test.

Antialiasing filter 25, maximum vertical resolution in pixels for 16:9 video crop area

Antialiasing filter 25, maximum vertical resolution in pixels for 16:9 video crop area as a function of focal length and aperture.

Antialiasing filter 50, maximum vertical resolution in pixels for 16:9 video crop area

Antialiasing filter 50, maximum vertical resolution in pixels for 16:9 video crop area as a function of focal length and aperture.

Antialiasing filter 75, maximum vertical resolution in pixels for 16:9 video crop area

Antialiasing filter 75, maximum vertical resolution in pixels for 16:9 video crop area as a function of focal length and aperture.

The color coding is set to be greenish in the range of 950 to 1300 (useful range for anti-aliasing), going red for lower values (noticeable blur) and blue for higher values (limited antialiasing effect).

As it can be seen, the blur effect varies with the aperture and focal length in a consistent manner. Increasing the focal length or enlarging aperture size creates more blur. The blur follows some optical laws linked to the aperture and focal length, I believe  somewhat related to the depth of focus.

A strong filter, such as strength 75, is more suited for wide angles lenses, e.g. f=18mm @  f/5 gives good results. The weaker filter, 50 or 25, is more useful for longer focal lengths, or for wide angles with very large apertures.

Below are video clips which demonstrate the filter effect according to the different pixel resolutions levels in the chart. Recorded with Nikon d5100 and the kit lens 18mm, and autofocus. Watch in 1080 HD, full-screen, and notice the effects on the brick wall!

~2160 px, No filter, f=18 mm. Strong moire!

~1650 px: Filter 75, f=18 mm, f/7.1. Strong moiré, no filter effect.

~1490 px: Filter 75, f=18 mm, f/6.3. Reduced but still a lot of moiré.

~1350 px: Filter 75, f=18 mm, f/5.6. Reduced but still visible moiré.

~1200 px: Filter 75, f=18 mm, f/5. Hardly visible moiré.

~1100 px: Filter 75, f=18 mm, f/4.5. No moiré, sligthly more blurred?

~990 px: Filter 75, f=18 mm, f/4. No moiré, some blur effect

~930 px: Filter 75, f=18 mm, f/3.5. No moiré, but clear blur.

All videos are uploaded raw to youtube to preserve original quality.

Wow, this brick wall really shows some funny artifacts when no filter is applied. Increasing the filter effect gradually eliminates the moiré, until the blur takes over. To get a 100% elimination of the moiré the image gets a bit blurry and loses details, while the for the weaker filters, the images appears more sparkling.  So, it is a matter of finding an optimal balance between acceptable moiré and sharpness level in the video for given video situations.  At about 1100 px there seems to be nearly 100% elimination of moiré while keeping a fair amount of sharpness. At the 1300 px level the sharpness is slightly higher, but moire is sneaking in from time to time. The improvement in image quality compared to the unfiltered version is still quite noticeable.  Going below the 950 px, makes the image becomes clearly soft and loses most of the HD appearance.

I hope the charts and the videos give an understanding of the nature behind moiré patterns.

More technical details

The test was made using a standard kit lens 18-55mm zoom lens and a 28 mm (2.8) Nikon lens with the Nikon d5100 and no UV filter. The sharpness was measured from still images of a star chart. The results applies directly for a camera system with the same chip size as Nikon d5100 which is 16.2 MP DX-format.  If using a camera with a different chip size and pixel density the resolution values should be scaled accordingly. Generally, a larger sensor will need a stronger filter to get the same effect. More details on the camera settings and the filter method can be found in the previous post.

At the moment I have not tested the filter on any other camera models, especially I dont know how the filter performs on Canon cameras. There could potentially be conflicts if the cameras already use image processing methods to reduce aliasing.

About these ads

Responses

  1. [...] DSLR Antialiasing filter – Sharpness charts « Recordings of nature's Blog on September 25, 2012 at [...]

  2. [...] filter frame (size 49, 52, 55, 58, 62 and 67 mm). See more info about the filters in earlier posts, here and here.  A small-scale piecewise assembly has been established. If interested, please contact me [...]

  3. […] with annotations. More testing: http://recordingsofnature.wordpress.com/2012/09/25/dslr-antialiasing-filter-sharpness-charts/Get the filter here: […]

  4. […] with annotations. More testing: http://recordingsofnature.wordpress.com/2012/09/25/dslr-antialiasing-filter-sharpness-charts/Get the filter here: […]


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

Follow

Get every new post delivered to your Inbox.

Join 49 other followers

%d bloggers like this: