Reduced OLP filtering on Sony Alpha 55 or why OLP is needed?!

[ErikKaffehr]

Hi,

I recently shot a comparison of my Sony Alpha 900 and Sony Alpha 55SLT using Norman Koren's MTF test target. The targets were shot with a 50 mm lens at about 5 meters.

Interestingly, both images show a significant amount of false resolution. Both also show some color moiré, the SLT a lot while the Alpha 900 has some. Note that the text below the resolution chart has a lot of color artifacts on the Sony Alpha SLT.

The sample here is shown 2:1 (double actual) pixels for easy viewing. The images are actual pixels shot from the same tripod position. The A55 SLT has a smaller pitch (4.77 microns) while the Alpha 900 has 5.9 micron pitch. So the Sony Alpha resolves higher in this setup, because the smaller pitch, while the Sonya Alpha 900, a full format camera, would have a wider field of view. In real world the Alpha would resolve more detail than the Alpha 55 due to larger sensor size, but that factor has been eliminated in this test.

I have marked the calculated Nyquist limit. It seems that the setup resolves beyond the calculated limit, probably due to incorrect setup (focal lenght longer than 50 mm and/or distance shorter then 5 m).


Best regards
Erik

[Bart_van_der_Wolf]

I have marked the calculated Nyquist limit. It seems that the setup resolves beyond the calculated limit, probably due to incorrect setup (focal lenght longer than 50 mm and/or distance shorter then 5 m).

Hi Erik,

That's one of the disadvantages of such a test target, it's sensitive to absolute distance and orientation (alignment with the sensel grid). It's why I prefer to use a target like the one I made (because it's better, and not because I made it ), because it is insensitive to things like absolute distance, or even magnification due to focal length (which is rarely the exact same as indicated on the lens), or sensel density. It also reveals small errors in shooting technique, e.g. mirror shake or heavy trafic in the street. It is even possible to distinguish resolution differences at different orientations in a single shot.

Some versions of Imatest can also produce such a radial sinusoidal grating star target, since Norman Koren added it when I suggested it to him. There is even an ISO test procedure that uses it, but it is also easy to visually evaluate some of the effects.

Cheers,
Bart

[ErikKaffehr]

Hi Bart,

Thanks for your advice on test targets. I'll do some testing with that target, too.

In my view the Norman Koren test target is actually better to illustrate the phenomena I wanted to check, namely if aliasing is visible on the A55 SLT? What I would the test target clearly shows is that we have both false resolution and color artifacts on both images but much more pronounced on the SLT 55.

Another observation I have made is that the Alpha 900 doesn't resolve the fine text under test chart while the SLT has pretty bad artifacts.

Best regards
Erik

Hi Erik,

That's one of the disadvantages of such a test target, it's sensitive to absolute distance and orientation (alignment with the sensel grid). It's why I prefer to use a target like the one I made (because it's better, and not because I made it ), because it is insensitive to things like absolute distance, or even magnification due to focal length (which is rarely the exact same as indicated on the lens), or sensel density. It also reveals small errors in shooting technique, e.g. mirror shake or heavy trafic in the street. It is even possible to distinguish resolution differences at different orientations in a single shot.

Some versions of Imatest can also produce such a radial sinusoidal grating star target, since Norman Koren added it when I suggested it to him. There is even an ISO test procedure that uses it, but it is also easy to visually evaluate some of the effects.

Cheers,
Bart

[Fine_Art]

I don't see the point of a test pattern with vertical bars or horizontal bars. This puts your sensor at its best by lining up pixel rows with bars. Who goes out to take pictures of bars? You need a test pattern that shows how the camera will respond to the real world. I use the Edmunds chart.
http://www.edmundoptics.com/products/displayproduct.cfm?productid=1665

This has different colors to test the weaker red, blue, yellow capacity. The bars are displayed in several directions by putting the USAF 1951 test pattern at angles.

Added: if you look on the 3rd tab you see their writeup on test charts for optical systems. All they do is optics.

[ErikKaffehr]

Hi,

The discussion is about OLP filtering. The reason I have chosen the target is pretty much that it would show moiré artifacts and it does indeed. Please note that the text under the mages also shows the same artifacts, so the test is certainly relevant for reproducing fine detail like text, stamps or Dollar bills.

Best regards
Erik

I don't see the point of a test pattern with vertical bars or horizontal bars. This puts your sensor at its best by lining up pixel rows with bars. Who goes out to take pictures of bars? You need a test pattern that shows how the camera will respond to the real world. I use the Edmunds chart.
http://www.edmundoptics.com/products/displayproduct.cfm?productid=1665

This has different colors to test the weaker red, blue, yellow capacity. The bars are displayed in several directions by putting the USAF 1951 test pattern at angles.

Added: if you look on the 3rd tab you see their writeup on test charts for optical systems. All they do is optics.

[Fine_Art]

If you are getting results beyond nyquist there is clearly something wrong with the test. Try rotating it 20 degrees.

[Fine_Art]

With the Sony 100 macro on the A55 at 30 ft I am resolving 22.56 seconds of arc or 83 lp/mm. Good primes should all be about the same.

Added: Or comparing it to human vision "Snellen defined “standard vision” as the ability to recognize one of his optotypes when it subtended 5 minutes of arc. Thus the optotype can only be recognized if the person viewing it can discriminate a spatial pattern separated by a visual angle of 1 minute of arc." -wikipedia

[Bart_van_der_Wolf]

Hi Bart,

Thanks for your advice on test targets. I'll do some testing with that target, too.

In my view the Norman Koren test target is actually better to illustrate the phenomena I wanted to check, namely if aliasing is visible on the A55 SLT? What I would the test target clearly shows is that we have both false resolution and color artifacts on both images but much more pronounced on the SLT 55.

I fully agree that it is useful to show real resolution, and make a distinction between that and ambiguous resolution (aliasing, mimicking as real resolution). Since there cannot be real resolution at / or beyond the Nyquist frequency resolution, my target allows to set that boundary (irrespective of actual shooting distance, or magification factor) by drawing a circle in the captured image at 92 pixels diameter (91.7 pixels = Nyquist), and determine that distinction at 'any' angle (not just at haphazard alignment with the sensel grid). It also works with a film capture of the target, and a subsequent scan (where higher sampling density of the same frame will reveal higher system resolution).

Cheers,
Bart

[Fine_Art]

Ive found a lot of fake resolution is from the raw converter "guessing" at line patterns, not the camera. The Sony IDC which many people hate for the limited features seems to do a smoothing +USM under the hood. Maybe it was Erik that pointed it out to me. On the plus side Ive never seen fake resolution from it. I have seen moire on plastic fabrics. Other converters ive tried have shown fake resolution on the same RAWs, pixel level differences that I knew were not there on the real thing. Maybe the beyond nyquist stuff is from an agressive RAW converter?

[ErikKaffehr]

Hi,

Two points:

1) Both cameras show spurious resolution on the right side, but it is pretty pronounced on the Alpha 55 than on Alpha 900. I guess it's because of weaker AA-filtering on the A55

2) The A55 show color artifacts on the text below the resolution chart, the Alpha 900 does not.

The images were converted in Lightroom and the fake resolution is not dependent on sharpening. I'll reshoot the comparison with Bart's test target when I have time.

Best regards
Erik

Ive found a lot of fake resolution is from the raw converter "guessing" at line patterns, not the camera. The Sony IDC which many people hate for the limited features seems to do a smoothing +USM under the hood. Maybe it was Erik that pointed it out to me. On the plus side Ive never seen fake resolution from it. I have seen moire on plastic fabrics. Other converters ive tried have shown fake resolution on the same RAWs, pixel level differences that I knew were not there on the real thing. Maybe the beyond nyquist stuff is from an agressive RAW converter?

[Fine_Art]

How's about using the standard ISO 12233 chart. If you are off the scale rotate it or put 2 together backing the camera further away.

http://www.graphics.cornell.edu/~westin/misc/res-chart.html

Bart's star chart is good, the only problem is measuring how far in you resolve. How do you know the line thickness at that point?

I use the edmund's test chart for buying a new lens. I know what im getting out of the camera system. The ISO chart does the same thing saving you $18 plus shipping. Normally the ISO chart costs you over $100. The link under the picture of it lets you print your own.

[Fine_Art]

I tried printing that out at 13x19. The smallest lines still go wonky.

Here is the Edmunds chart shot. The instructions are readable. The jpg is too big to attach here.

http://www.yourfilelink.com/get.php?fid=717453
This was shot at 1992 mm (Edit: from wall to camera mount) on A55 with 50 Macro ISO100 f5.6. The tape measure in the pic is leaning a bit from sitting on the floor, pointing in the air.

[Bart_van_der_Wolf]

How's about using the standard ISO 12233 chart. If you are off the scale rotate it or put 2 together backing the camera further away.

http://www.graphics.cornell.edu/~westin/misc/res-chart.html

The ISO 12233 chart needs to be shot at a very specific distance for the numerical indications to make sense, and they are not very well suited for software analysis. The hyperbolic 9 line targets on that chart are mostly intended for a visual impression. The only parts that can be used for a detailed (MTF) analysis are the (slightly) slanted edges that can be found at a few places in the chart.

Bart's star chart is good, the only problem is measuring how far in you resolve. How do you know the line thickness at that point?

It's quite easy, the resolution in cycles/mm = (144 / Pi) / diameter of the blurred center, measured in millimeters (or pixels times sensel pitch). It's a very close approximation which can usually only be improved in the second fractional decimal by using a more complex formula.

Cheers,
Bart

[Fine_Art]

The ISO 12233 chart needs to be shot at a very specific distance for the numerical indications to make sense, and they are not very well suited for software analysis. The hyperbolic 9 line targets on that chart are mostly intended for a visual impression. The only parts that can be used for a detailed (MTF) analysis are the (slightly) slanted edges that can be found at a few places in the chart.

It's quite easy, the resolution in cycles/mm = (144 / Pi) / diameter of the blurred center, measured in millimeters (or pixels times sensel pitch). It's a very close approximation which can usually only be improved in the second fractional decimal by using a more complex formula.

Cheers,
Bart

So you need a length overlay on it. You can't put a tape measure on your screen, you don't know how much the image has been scaled.

You can measure the height of the image, measure again on your screen then use the ratios to decide on the mm pitch when it went blurry. You can do the same ratio using the ISO chart, or the edmunds chart. The edmunds chart just happens to have the mm pitch of all the lines labeled.

[Hans van Driest]

If the MTF testchart was used for a Leica M9 or a medium format back, the results would be the same, and the quality of the camera would be praised. The same property that results in undesired aliasing artifacts, in the shown result, also result in high micro contrast in real life pictures. In a landscape, false detail can be very pleasing, suggesting detail where the sensor resolution is insufficient.
I often use a NEX 5n for landscape panorama’s because of the absence of an AA filter. And I must say that visual artifacts are hardly ever present. The positive aspects often are, when lenses are used that have sufficient resolution.

Hans

[bjanes]

I don't see the point of a test pattern with vertical bars or horizontal bars. This puts your sensor at its best by lining up pixel rows with bars. Who goes out to take pictures of bars? You need a test pattern that shows how the camera will respond to the real world. I use the Edmunds chart.
http://www.edmundoptics.com/products/displayproduct.cfm?productid=1665

This has different colors to test the weaker red, blue, yellow capacity. The bars are displayed in several directions by putting the USAF 1951 test pattern at angles.

Added: if you look on the 3rd tab you see their writeup on test charts for optical systems. All they do is optics.

The USAF target employed in the Edmund chart has been in use for over 50 years, but its limitations are discussed by Norman Koren. The high contrast bars are most suited for estimating high resolution at a low MTF (around 10%), but for practical photography one is more interested in MTF at lower resolution. For MTF evaluation, a sinusoidal pattern is more useful. Perhaps Bart can elaborate.

Bill

[Bart_van_der_Wolf]

The USAF target employed in the Edmund chart has been in use for over 50 years, but its limitations are discussed by Norman Koren. The high contrast bars are most suited for estimating high resolution at a low MTF (around 10%), but for practical photography one is more interested in MTF at lower resolution. For MTF evaluation, a sinusoidal pattern is more useful. Perhaps Bart can elaborate.

Hi Bill,

Yes, the USAF target is originally intended to be used with film (shooting from high flight altitude, and determining the resolution that could be achieved for subjects on the ground). The high spatial frequency edges of the bars make the target very sensitive to alignment with a regular sensel grid, and is therefore less suited for digital sensors. It is a high contrast target, and doesn't really allow to be used for an MTF analysis, where the contrast reduces with increasing spatial resolution (because the sharp edges are at the resolution limit, but the bars are not, it's a mixed signal).

For digital sensors, sinusoidal gratings and slanted edges are the way to go, as evidenced by several ISO standard procedures for testing resolution (of scanners and cameras).

Cheers,
Bart