AF101 BBC/EBU test results posted

[Lpowell]

The one thing that continues to confuse me is simply the discrepancies in advertised [800-line] TVL compared to what these tests are producing.

I think a careful reading of Alan Roberts' report and a close look at his 1080p Zone Plate photos will cast some light on the nature of these discrepancies (be sure to view the PDF file at 100% magnification):

http://thebrownings.name/WHP034/pdf/WHP034-ADD66_rev_2_Panasonic_AF101.pdf

Roberts has this to say about the luma Zone Plate on page 9 of the report (my emphasis):

Resolution is clean up to 63% of 1920x1080, where there are low-level null zones visible. This means that the image has clean resolution only up to 1210x680, which is not good, given the strength of the aliases.

If you examine the Zone Plate in detail, I think you'll see exactly what Roberts means by "clean resolution". It's not the case that the AF100's image details become indistinct or blurry above 680 lines. To my eyes, the high resolution details remain sharp at well over 900 lines, however, they are unfortunately contaminated by equally prominent aliased details that extend throughout much of the resolution range. Pinpointing exactly where the aliasing could be said to start is really a judgment call, which Roberts makes as 680 lines. In a case like this, Panasonic's technical reading of the Zone Plate might well vary from Roberts', as may yours or mine.

I think the reality is not that the AF100 is lacking in sharpness or FHD image resolution. The dilemma is that its sharpness, both measured and perceived, is apparently a roughly equal mixture of genuine 1080p resolution and super-high-resolution artifacts that are aliased into the image as well. Roberts makes this telling observation that backs up that conclusion (my emphasis):

...It is also clear, from the double-frequency pattern in the middle [of the Zone Plate photo], that the lens is passing frequencies up to nearly twice that of 1920x1080, since the vertical spatial alias reappears above the sensor resolution limit. This is an indication that there is either no optical low-pass filter, or that it is inadequate.

If this is in fact the case, then what the AF100 really needs is a damn good low-pass filter in front of the image sensor to block the super-high-resolution details that are contaminating the image as artificially sharp, aliasing artifacts. We would then be able to test and observe the camera's true FHD resolution.

 

[Barry_Green]

What difference does the use of a binary ZP vs. a sinusoidal ZP make in these experiments? The one used in the Wringer appears to be a binary ZP.

Can't imagine why you'd want a sinusoidal zone plate for a res chart... you want the clearest, most distinct transition from black to white (or from color to color, or whatever) that you can get. Blurring the transition from black to white (as is done in a sinusoidal plate) would just complicate matters terribly.

 

[Steve Kahn]

One question though - It seems he tested with these settings:

detail 0
v detail -2
corning 0

Meaning he's actually adding sharpening which could add aliasing and skew the test, no?

To do a resolution test wouldn't one want to remove ALL artificial sharpening?

 

[artyswiss]

Boom.

 

[nugat]

One question though - It seems he tested with these settings:

detail 0
v detail -2
corning 0

Meaning he's actually adding sharpening which could add aliasing and skew the test, no?

To do a resolution test wouldn't one want to remove ALL artificial sharpening?

So you assume that "0" means adding sharpening? And what if "0" is here a real "0" and "-" means really negative detail?
That's what the author conjectures in the white paper on page 10.

"The detail settings did not significantly improve the situation,
the aliases dominated at all settings, however, setting vertical
detail to -2 did look a little better. It seems possible that the
detail level controls in this camera actually reduce detail with
negative settings, an usual but welcome feature."

Or shall we follow the method of taking from Mr Roberts what's convenient for us and
denying what's not? No? Boom.

 

[Steve Kahn]

A setting of 0 (ZERO) most assuredly IS adding artificial sharpening. That is well documented.

-7 is no sharpening.

Also, no where did he document testing all settings possible. In fact how practical or even possible in the limited scope of a broad camera test?

I think this resolution test is flawed when artifical sharpening could easily lead to aliasing and therefore loss of resolution. I feel the author didn't even know that -7 was no added sharpening otherwise that would have been the natural setting one would think.

 

[Mike Harvey]

I've been shooting like crazy since getting my AF-100 the past 2 months, and clients seemed thrilled. I just wrapped 10 days on a tv pilot (news-magazine style) and the director and editor were thrilled with the shoot. Other clients seem thrilled too. I don't necessarily work in a world where clients need stuff from a RED or an Alexa, but I think this camera delivers nice images and it makes people happy. Not to toot my horn, but I've billed over $20k in shoots since getting this camera, so, yes, I think that it's worth it. And I may buy something else later this year, but it will have gotten good use until then.

At the end of the day, this is the test that truly matters. Will this camera produce images that will make clients happy and make you money. Ben's statement makes me want to buy this camera (even after reading this review for the BBC) that much more.

 

[nugat]

I think it's worth raising it with the author with proper proofs. He is a live person, a nice gentleman and answers when asked.
No need to do exegesis on his writings.
Can you link the "sharpness" issue here as well?

 

[Lpowell]

I think this resolution test is flawed when artifical sharpening could easily lead to aliasing and therefore loss of resolution.

The camera's sharpening filter acts on the image sensor's RGBG output after aliasing has already been produced by the sensor's digital sampling of the light focused through the lens. While a properly engineered sharpening filter may emphasize aliased artifacts that already exist in an image, it will not create any additional aliasing artifacts itself.

In this case, the prominent aliasing artifacts seen in the Zone Plates were clearly generated by super-high-resolution details that were not adequately blocked by the camera's optical low-pass filter. The sharpening filter is not the culprit as it cannot produce this type of aliasing effect.

 

[J Michael]

And that's the reason I asked. The chart shown on page 10 of the report appears to be a sinusoidal ZP whereas the Wringer appears to be binary. Google seems to turn up more references to sinusoidal than binary for sensor resolution tests.

Can't imagine why you'd want a sinusoidal zone plate for a res chart... you want the clearest, most distinct transition from black to white (or from color to color, or whatever) that you can get. Blurring the transition from black to white (as is done in a sinusoidal plate) would just complicate matters terribly.

 

[Lpowell]

Can't imagine why you'd want a sinusoidal zone plate for a res chart... you want the clearest, most distinct transition from black to white (or from color to color, or whatever) that you can get. Blurring the transition from black to white (as is done in a sinusoidal plate) would just complicate matters terribly.

Actually, due to the mathematics of digital image sampling, the reason a sharp-edged transition is not preferable in a Zone Plate is because it would in fact complicate matters in the spatial frequency domain that's being measured. The concentric rings in the chart are designed to test the camera's response to individual degrees of resolution. If you sharpen those sinusoidal edges, you would introduce additional degrees of resolution at higher spatial frequencies than you're attempting to measure.

In the charts used by Roberts, the outermost circle is calibrated at a resolution of exactly 1080 lines, which is the highest resolution the chart is designed to test. It would create superfluous details if the chart was rendered at a higher resolution than that.

 

[Steve Kahn]

The camera's sharpening filter acts on the image sensor's RGBG output after aliasing has already been produced by the sensor's digital sampling of the light focused through the lens. While a properly engineered sharpening filter may emphasize aliased artifacts that already exist in an image, it will not create any additional aliasing artifacts itself.

In this case, the prominent aliasing artifacts seen in the Zone Plates were clearly generated by super-high-resolution details that were not adequately blocked by the camera's optical low-pass filter. The sharpening filter is not the culprit as it cannot produce this type of aliasing effect.

This may or may not be true. I am not a camera technician nor an optical engineer. But I do have a b.s. in physics and I can tell you it is plainly not good scientific methodology to do a test with extraneious variables if they can possibly be removed. And artificial sharpening could easily have been removed and should have been removed before testing.

V detail was lowered to -2 and improved things so there must be some effect of sharpening which seems to deflate your conjecture.

I wonder if -7 detail, -7 v detail was even tested.

 

[Lpowell]

...it is plainly not good scientific method to do a test with extraneious variables if they can be removed. And artificial sharpening can easily be removed and should have been removed before testing.

The camera's sharpening filter cannot be removed because it is an integral part of the demosaicing algorithms used to calculate RGB pixel values from the RGBG Bayer Filter output of the image sensor. All digital cameras contain built-in sharpening filters to fine tune the spatial frequency response of their image sensors. Your speculation that the sharpening filter is "extraneous", "artificial", or "could easily lead to aliasing" is technically unfounded.

 

[BobbyMurcerFan]

Posted in wrong thread.

 

[Matt Gottshalk]

All I know is my clients love the images, and it makes me money.

Anything else is extraneous.

 

[pulpfiction007]

The camera's sharpening filter cannot be removed because it is an integral part of the demosaicing algorithms used to calculate RGB pixel values from the RGBG Bayer Filter output of the image sensor. All digital cameras contain built-in sharpening filters to fine tune the spatial frequency response of their image sensors. Your speculation that the sharpening filter is "extraneous", "artificial", or "could easily lead to aliasing" is technically unfounded.

If I'm not mistaken I don't think Steve ever advocated removing the sharpening filter. I think all he is getting at is to test with: detail -7 v. detail -7 sharpening -7. Maybe this would help reduce the emphasis of the aliasing. Even if this doesen't improve the test results, but improves the look as Roberts said v detail at -2 did , it would be worth noting for certain types of shots.
BTW...lpowell - I'm not saying you are wrong in any way about the technical issues, as I will defer to you and others for that data...just that maybe Steve's point about testing with different parameters could yield interesting results.

 

[Stephen Mick]

Let's not make this all personal. if some people are put off by the findings of the BBC, they have every right to be. If others are perfectly happy with the footage they're getting from the AF-100, that's great. But we can, I think, ask questions, debate findings, and respect each other's positions and opinions without the tone going negative.

 

[Steve Kahn]

If I'm not mistaken I don't think Steve ever advocated removing the sharpening filter. I think all he is getting at is to test with: detail -7 v. detail -7 sharpening -7. Maybe this would help reduce the EMPHASIS of the aliasing. Even if this doesen't improve the test results, but improves the look as Roberts said v detail at -2 did , it would be worth noting for certain types of shots.

Yes. Sorry if I was not clear.

I am pointing out that on page 3 of the report the user settings employed were:

detail 0
v detail -2
corning 0

I don't quite understand using these settings in a resolution test. (again I'm not a technician, please tell me if I'm mistaken and why).

Intuitively -7 for these three settings would have made the most sense to eliminate aliasing.

 

[Lpowell]

I don't quite understand using these settings in a resolution test. (again I'm not a technician, please tell me if I'm mistaken and why).

Intuitively -7 for these three settings would have made the most sense to eliminate aliasing.

Fundamentally, a digital sharpening filter can neither create nor eliminate aliasing. Aliased artifacts are produced at the point where the light focused by the lens is digitally sampled by the image sensor. If the light patterns contain spatial details that are finer than the physical resolution of the sensor, many of these super-high-resolution details will be folded into the image as artificially sharp edge artifacts. That is why an effective optical low pass filter is needed to block image details that are higher in resolution than the sensor is designed to accurately detect.

Once aliased artifacts are imprinted into an image, there is no filtering algorithm that can selectively eliminate them. While the camera's sharpening filter can be dialed down to the minimum in an effort to conceal the aliased artifacts, that will also suppress genuine image details as well. The result will be an image with excessively smoothed edges that nevertheless still contains aliased artifacts. The intractability of this problem is often painfully apparent in dealing with the color-fringed aliases produced by the Canon 5D Mark II.

A good introduction to the physics used in measuring the spatial frequency response of digital image sensors can be found here:
http://www.imatest.com/docs/sharpness.html

 

[Faust]

Fundamentally, a digital sharpening filter can neither create nor eliminate aliasing. Aliased artifacts are produced at the point where the light focused by the lens is digitally sampled by the image sensor. If the light patterns contain spatial details that are finer than the physical resolution of the sensor, many of these super-high-resolution details will be folded into the image as artificially sharp edge artifacts. That is why an effective optical low pass filter is needed to block image details that are higher in resolution than the sensor is designed to accurately detect.

Once aliased artifacts are imprinted into an image, there is no filtering algorithm that can selectively eliminate them. While the camera's sharpening filter can be dialed down to the minimum in an effort to conceal the aliased artifacts, that will also suppress genuine image details as well. The result will be an image with excessively smoothed edges that nevertheless still contains aliased artifacts. The intractability of this problem is often painfully apparent in dealing with the color-fringed aliases produced by the Canon 5D Mark II.

I couldn't agree more.