Filter Test Parameters - suggestions needed

[rpcrowe]

We all know that top-line filters will not degrade imagery like "el-cheapo" brand filters might. But, just how much will the difference be?

I have never seen a test between top-line filters and the cheaper variety probably because folks seldom purchase two grades of filters. I have the opportunity to compare less expensive filters with better grade filters now.

I received a set of three 77mm Pro-Optic filters from Adorama, when I purchased my 100-400mm f/4.5-5.6L IS II lens, as a freebie at no additional cost. These are CPL, UV and 2x ND filters
http://www.amazon.com/ProOptic-Essen.../dp/B0077G7JUE

I have the same types of filters manufactured by B+W. The 77mm Pro-Optic filter kit runs $48.00 USD on Amazon USA. While there are certainly cheaper filters on the market, a comparable group of B&W 77mm filters would total $200 USD or more on Amazon; a 4x difference...

The B&W filters image quality should significantly beat the IQ of the Pro-Optic filters. But this will give us an idea of just how much better the expensive filters are than the less expensive models.

I plan to shoot my mannequin head in front of a brick wall at 200mm and f/8 with my new 100-400L II lens tripod mounted. I will include a WhiBal card and shoot in RAW with AWB.

First with no filter and then with the B+W and the Pro-Optic filters added.

I will open the image into Adobe Camera RAW. I will not sharpen the image – even though my normal workflow is to use the NIK RAW Presharpener as I open the RAW image.

For images to be posted on the Internet, I usually reduce them to 72 pixels per inch, saving them to JPEG and then upload them to www.smugmug.com. I then link to my smugmug image to post on the internet…

I expect to post them with no cropping.

I am not sure if I should shoot this test with my 7D Nark I or my 5D Mark II…

Are there any other things that I should be concerned with as I prepare to do the testing.

We "might" not see the difference in the images as posted. However, I will also give my opinion on the looks of the different filter shots in prints...

[DanK]

There is a lot in between $48 filter sets and B+W... BTW, lenstip.com has tests of filters, but I doubt they have any that inexpensive. (I haven't looked in a long time.) That's how I found out about Marumi, which is now my default brand.

I have a hunch you won't see much in terms of resolution. First, the resolution of computer screens is very low, which will make it hard to see any degradation of accutance. Second, I'm assuming you will have lights behind you. Cheap filters are more likely to cause a problem (ghosting, flare) with light to the side or in front. I did a test of a Hoya UV filter vs. no filter once in response to the filter wars online, and it showed no difference, but I realized later that it not a good test for the first of these reasons. The one aspect of filter quality that I think can be easily demonstrated online is color cast, which might show up if your set includes a CPL or an ND.

To get a better comparison, I would suggest two things:

1. Upload the files at maximum size to a smugmug gallery set to allow viewing at original size. Then post the links in addition to embedding them. This will allow people to access the full-size images if they want.

2. Try lighting at different angles.

3. Set the camera to a fixed kelvin temperature, and make sure that Lightroom has correctly read the exif and imposed that temperature in rendering the raw file. If you use AWB, the camera will try to offset any color cast.

[Dave Humphries]

Hi Richard,

Do not 'reduce' them to 72 dpi - the concept is completely meaningless anyway for on-line display - and you just risk changing something else in the process.

The files will be whatever pixel dimensions the camera's sensor is, just leave them alone and upload (and display) at full size.

I agree with Dan's advice.

What would I expect of a cheap filter?
a) Flare due to lower quality, or at worst, no multi-coating on one or both sides of filter - this might reveal itself as an overall 'fog' or lack of contrast - think how best to see this and ensure there are things in shot to make it visible and not mask it - and/or reflections of any light sources in shot - think how best to see these and ensure there are things in shot to make it visible and not mask them
b) Colour cast, with UV or ND particularly
c) Unlikely to see resolution reduction* (unless really bad)
d) A thread that is easily damaged or sticks on your lens

* I cannot envisage that a brick wall will be suitable for resolution checking anyway (unless it is really bad) - think about MTF test charts and how different they are to a brick wall.

I'm sure there are more things ...

One final thought - is testing worth the effort?
Especially if you are not intending to use them.
They cost nothing, throw or (knowing you) give them away.

[Manfred M]

Richard - a decent lab-like setting with standardized test targets would be a place to start. Unless there is some obvious visible flaw in the glass, the type of testing you are planning to do are not going to be meaningful.

In a perfect world, I would test for the amount of light that is transmitted and reflected by the filter, any colour casts, resolution. This would required some specialized equipment and with the possible exception of the resolution test, I can't see a camera being involved an any of the testing.

[DanK]

Manfred,

You are a pessimist!

A color cast might be easy to pick up with the procedure I described. Here is an example. I tested several ND filters by shooting a wall with a print tacked to it, using a fixed temperature (which happened to be a bit off neutral, but since the point was a comparison, it didn't matter). The shots were identical except for changes in shutter speed to compensate for the ND. I did no editing; I just read them into LR and checked the color temperature. Here is the shot with no filter:



And here is the filter that that had the most noticeable color cast:



Not extreme, but certainly noticeable, and different from my other two, which showed only very slight differences from the no-filter shot. On the other hand, if Richard is testing UV filters, I doubt any difference would be large enough to be apparent. Perhaps true of CPLs too.

I also suspect that Richard would be able to get noticeable differences in flare, if he illuminated from in front of the camera. However, I agree with you that many of the quality differences would only be apparent with more rigorous testing.

Dan

[Ken MT]

I am not sure if I should shoot this test with my 7D Nark I or my 5D Mark II…
...

What's a Nark I

1: To be a nark is to be someone that will report any misbehaving to a higher authority like the police, parents and teachers.
2: To nark, or to nark someone out is the action of reporting the misbehavior; also called narking.
3: To be narked is, obviously, the past tense of narking.

[Ken MT]

Slightly old article, but worth reading

http://www.lenstip.com/113.1-article...ters_test.html

But more recent for polarizers

http://www.lenstip.com/139.25-articl...d_summary.html

[Manfred M]

Manfred,

You are a pessimist!

Not a pessimist. Too many years of designing tests and doing benchmarks.

There are two test approaches that tend to be in use. The first is the one I proposed as being my preferred approach which means testing against an absolute standard. The second is a comparative method, which is what you are doing, i.e. comparing. The issue with a comparative test is always with the assumption that one of the test subjects is good enough to be the benchmark. If you make a mistake there, the test becomes meaningless. That is something that tends to happen all too often.

[Tringa]

It would be interesting to see the real world results, Richard. I agree reduction of the photos for the Web is likely to make any differences impossible to see and that, if possible, taking a number of shots with a light source to the side but in front of the lens would be useful too. Hopefully there will be a point where the cheaper filters show flare but the more expensive ones do not.

Dave

[DanK]

Not a pessimist. Too many years of designing tests and doing benchmarks.

There are two test approaches that tend to be in use. The first is the one I proposed as being my preferred approach which means testing against an absolute standard. The second is a comparative method, which is what you are doing, i.e. comparing. The issue with a comparative test is always with the assumption that one of the test subjects is good enough to be the benchmark. If you make a mistake there, the test becomes meaningless. That is something that tends to happen all too often.

We're going off the original topic here, but this is interesting (at least to me), so hopefully no one will mind continuing this.

The key in my mind is this sentence:

The issue with a comparative test is always with the assumption that one of the test subjects is good enough to be the benchmark.

Not necessarily. It just requires the assumption that differencing two measures provides an unbiased estimator. If that is true, a comparative measure will be noisier, but that noise can be usually be quantified and addressed. A huge portion of real-world statistical analysis follows this model. E.g., most tests of the drugs necessarily entail comparison of two or more groups, as there usually no absolute measure of the condition in question, and even when there is, that measure is not necessarily stable (people get better or sicker).

In this case, however, there actually was a benchmark. My question was, "to what degree, if any, would adding one of my ND filters change color compared with using the same equipment with no filter at all. I did worry about noise, so I repeated the test a few times, under somewhat different conditions, and obtained essentially identical results.

I suspect our different views of this come from what we both did for a living. I take it you are an engineer, and I can see that in many engineering applications, a comparative test would be a sloppy compromise, given the specific questions you have to address. I make my living doing statistical analysis of social phenomena, where almost all rigorous work is necessarily comparative.

None of which directly helps Richard, of course. But interesting to me, at least.

My apologies for the digression.