Sony EX1 Resolution Tests - Imatest MTF50

[Tom Roper]

I just finished the MTF50 tests on the Sony EX1 camcorder. This is a camera I own, and the procedure is identical to the method I used previously when testing the Canon XH-A1 and HV10 HERE , so the results are directly comparable.

The camera scored very well particularly at HQ1080p24/p30, but is about the same resolution wise at 35 mbps VBR, HQ1080i60 as the Canon XH-A1 at HDV 25 mbps 1440i60.

The resolution charts were shot from a tripod at a distance 7 feet from the lens. Uncompressed bitmaps were extracted from the output streams, and used as the input to the Imatest software. The results are normalized to a standard 2 pixel sharpening radius.

The camera image stabilizer was off. White balance was ATW, Gamma was STD3, Detail on, Gain -3, Iris F4.8, Shutter Off.

Summary:

HQ 1080P30 Horizontal - 931 LW/PH (line widths per picture height, i.e. "tv lines.")
HQ 1080P30 Vertical - 1109 LW/PH

HQ 1080i60 Horizontal - 888 LW/PH
HQ 1080i60 Vertical - 662 LW/PH

*******************************************

EX1Horz30p
EX1Vert30p
EX1Horz60i
EX1Vert60i

[Paul Chiu]

tom,
what does this mean? EX1 you own as line resolving as the other cams at the same price range? higher or lower? also, did you run tests under different lighting conditions?

thanks

paul

[Dominik Seibold]

You should really redo the test with detail-off. The arbitrary standard-detail-settings aren't very interesting but have a huge effect on the test. The raw output with detail-off is much more interesting. I would not be surprised at a better result with detail-off.

[Dominik Seibold]

I just have to pronounce how amazed I am about this ex1. Today I shot a rap-musicvideo and again the picture-quality this cam delivers with detail-off just blew me away. Here's a grab:
http://www.dominik.ws/sas.png
(cine3, detail-off, matrix-off, black/black-gamma/gamma-level 0)
If there weren't that rolling-shutter (which is visible while very shaky handheld-shots), this cam would just be PERFECT. I can't imagine any better (ok there could be less noise in dark areas, but what about film-grain ;) ). And this long-gob-mpeg2-format is very nice, because it saves a lot of memory compared to i-frames-only-formats and artifacts are almost invisible.

[Tom Roper]

Sharpening will increase the perception of resolution, and actually increase the measurement of it as well. The Imatest software takes this into account, and equalizes the result into a standard level of in-cam sharpening, i.e. a 2-pixel radius, so that cams with differing levels of applied sharpening can be compared on a level playing field. The detail setting "on," while somewhat maligned is really more of a preference because the amount of sharpening (as can be seen from the curves) is not out of line with other cams.

If you look at the edge profile in the upper half of the chart, you'll get higher numbers, I would characterize this as resolution at the point of extinction. MTF50 is a different way of looking at it. MTF50 asks what is the resolution at the point between black and white where the contrast equals 50%, or in other words, gray. If you eyeball a chart, you can still see some faint lines near the point of extinction. MTF50 gives a more conservative number. Both are valid ways of looking at it. The Imatest software just removes the human element from the measurement.

In any case, the numbers are really high for HQ 24p/30p. The numbers are actually very high for 60i compared to other cams I've measured.

Resolution is a criteria, not a hysteria. The EX1 produces fantastic images out of proportion for its price range.

I will do more tests if there is something to be learned, but already there is something learned...that 24/25/30p is the route for maximal spatial resolution, 60i is the way to get temporal resolution. There are a lot of beautiful images coming our way with both formats.

[Dominik Seibold]

So your software compensates more or less (it just doesn't know the exact kernel of that (variable) sharpening-filter) the sharpening. But isn't it still in principle more reasonable in a measurement-process to bypass everything in the signal-path which you don't want to measure? Also the detail-feature doesn't only sharpen, but also does noise-reduction, which is for sure a non-linear and hardly predictable thing.
I don't know that mtf50-standard, so could you explain how it is possible that a device with 1080 lines has a vertical resolution of 1109? I don't understand that "line widths per picture height". I guess the picture-height is 1080. But what does line-widths mean? Perhaps I understand it immediately, if you tell me the theoretical perfect values for a (progressive) 1920x1080-device.

[Steven Thomas]

Very true..

I'm far more interested in what an image looks like with artificial enhancement set to its off position.

Many of these sub $10K cameras do not look good at all. The EX1 is one of the few that holds up with detail OFF and produces a smooth natural high rez image.

[Paul Chiu]

Originally Posted by Dominik Seibold

..............
(cine3, detail-off, matrix-off, black/black-gamma/gamma-level 0)
.........

dominik,
were these settings after numerous tests you ran?
did you have image stablization on when you experienced the "rolling shutter"?

[Steven Thomas]

Originally Posted by Paul Chiu

dominik,
were these settings after numerous tests you ran?
did you have image stablization on when you experienced the "rolling shutter"?

??
I don't see where he wrote he experienced rolling shutters issue during his rez test.

[Dominik Seibold]

Originally Posted by Paul Chiu

were these settings after numerous tests you ran?

Yes. Cine3 emerged for me to be the default gamma for average dynamic scenes to get that cool a bit contrasty filmic look. In low-contrast-scenes I use cine1. In very contrasty scenes I would sometimes use cine4, but also then mostly cine3, because it looks a bit more punchy.
Detail-off looks pin-sharp and smooth at the same time, so I use it always.
The standard-matrix is very similar to matrix-off. Cinema-matrix reduces color-saturation. If I really would want that look, I would do it in the post (why replace information with noise while acquisition?). HighSat is unnatural colorful (and leads sometimes to chroma-clipping). So I just don't use the matrix-feature. (You can assume I'm longing for a raw-workflow ;) )
The black-level shifts the black-point. But why should I want to do that? If I shift it up black gets grey (when useful??). If I shift it down grey and black gets the same (this is useful, if there is no black, but in most cases there's something black). So I don't use it.
black-gamma is very useful to get more contrast (or decrease) into dark areas, so I use it sometimes.
gamma-level isn't very useful. Increasing it means to decrease the maximal recorded IRE (with cine-gammas). Decreasing it means to cut off light areas (like decreasing the white-point in Photoshops levels + decrease black-contrast). In any case I throw away picture-information, So I don't use the gamma-level.

Originally Posted by Paul Chiu

did you have image stablization on when you experienced the "rolling shutter"?

Watch this clip from 4:46 to 5:33:
http://www.youtube.com/watch?v=o9ChXxVV-ww
I used steadyshot, but my hand-movements aren't smooth enough (or steadyshot is too limited). You can see a bit of a jello-effect. Very subtle, but it's there.

[Tom Roper]

Originally Posted by Dominik Seibold

I don't know that mtf50-standard, so could you explain how it is possible that a device with 1080 lines has a vertical resolution of 1109? I don't understand that "line widths per picture height". I guess the picture-height is 1080. But what does line-widths mean? Perhaps I understand it immediately, if you tell me the theoretical perfect values for a (progressive) 1920x1080-device.

The screen grab of the resolution chart is a raster image. The Imatest software doesn't know the dimensions of that image. You draw a box around a wedge that is at an angle. You can see it on the chart.

Line widths per picture height is an expression of "tv lines." In the old days, tv lines were the number of cycles inside a circle whose diameter is the height of the vertical.

If you had a resolution of 1080 tv lines for both the vertical and horizontal dimensions, 1080 line width per picture height, there would be 1920 line widths across the full width of the image since the aspect ratio is 16x9.

We call them line "widths" because we only count the black ones or the white ones, not both.

In our example, the EX1 had 931 line widths in the horizontal dimension. If you consider across the full width of the image, 931x16/9 equals 1655 lines resolved to 50% contrast. That's not far from nyquist, (could be more if you counted lines near the extinction point) and as you noted in the vertical, it exceeds nyquist. That's just the way the software saw it, and it's not unusual. See it all the time on cams that use pixel shifting to artificially increase resolution beyond the number of physical pixels by interpolating them.

MTF50 means "modulation transfer function" or the number of cycles counted until the contrast decreases to 50%.

Perfect values for 1920 by 1080 would be 1080 tv lines in both directions, horizontal and vertical.

[Dominik Seibold]

Originally Posted by Tom Roper

Perfect values for 1920 by 1080 would be 1080 tv lines in both directions, horizontal and vertical.

So the ex1 is better than perfect? (1109 > 1080)

...
Btw, attaching a &fmt=18 to a youtube-url is cool. ;)

[Tom Roper]

[QUOTE=Dominik Seibold;886885]So the ex1 is better than perfect? (1109 > 1080)

The bottom of the chart shows one more number, for MTF50"P" which is 1043 LWPH. The MTF50P is commonly applied to video.

I was worried you were at first complaining the numbers were low :)

[Paul Chiu]

Originally Posted by Dominik Seibold

.........
I used steadyshot, but my hand-movements aren't smooth enough (or steadyshot is too limited). You can see a bit of a jello-effect. Very subtle, but it's there.

dominik,

the ex1 gave a gorgeous presentation of that night scene with just ambient "true" lighting. that "jello" effect during 446-533 is so minor i had to look twice. maybe if you had shot at 1080-60i, even that would have been fixed when viewed on a 1080p panel.

i'd found while filming kids at 24p, the ex1 just cannot follow them. i will shut off steadyshot and used 60i next time i shoot faster moving subjects.

..and thanks for sharing the cine-data. my 1st week with ex1 and so much, so much to learn.

paul

[Serena Steuart]

I assume Adam Wilt's tests have been noted?
http://provideocoalition.com/index.p..._camcorder/P3/