Canon XH A1 Resolution Imatest MTF50
 

Horizontal 730.0 lines
Vertical 659.5 lines

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

http://vsdrives.com/graphics/resolut...Canon_XHA1.PNG


http://vsdrives.com/graphics/resolut...Canon_XHA1.PNG

Notes about tests
 

I used my XH A1 and Imatest software to do the above SFR (spatial frequency response) measurements. The camera settings were done on MANUAL, 60i, 1/60th, F4.8 with OIS off. The grabs are made from bitmap captures on the native m2t file. No secondary processing was performed on the bitmap image. Since the native aspect is 1440 by 1080, the MTF50 result from the chart on the horizontal has to be normalized by a coefficient of 0.75x to account for display at 1920 x 1080.

The MTF50 (Modulation Transfer Frequency) test is the most conservative method, and the one that is most commonly used for testing lens/sensor combinations. The numbers above are normalized for standard 2 pixel sharpening, so as to make the results for cams using different amounts of sharpening comparable.

Unfortunately, numbers abound on the internet, but seldom are they accompanied by the testing method used, which in its simplest form could be somebody looking at a ISO12233 chart and eyeballing 800 x 800.

More objective and still valid resolution figures could be stated for this cam as 914H lines x 792V lines by reporting the result as MTF30, and the information would be valid as long as we weren't using it to compare to a cam measured using the MTF50. So the choice between MTF50 and MTF30 has less to do with which is more valid and more to do with the goal of producing consistent results that are comparable. Since MTF50 is the generally accepted and more conservative one, that's what I use here.

Tom,

Thank you very much for your efforts.

Just a general question here. Would it be correct for one to say that the potential for HD Resolution with this camera (given your measurements) would be 481,435 pixels from a possible max perfect goal of 2,073,600 (or 23.2% of High Definition's max resolution)?

Its always confused me why some ppl say "lines" and yet everything we see on LCD's is pixels and everything thats processes is in pixels. Especially resolution. i think it was due to interlaced TV signals being measured as lines and its just carried on over. But i'd like to be clarified just to make sure i'm on the right track.

Something tells me i'm not adding it up correctly as a little under a quarter resolution of HD is what the current "HD" camera's are actually doing - just doesnt sound right. But i remember researching cameras years ago and being really unhappy with the quality (or lack thereof) and ACTUAL resolution they were giving being far lower then even the recording format's potential.

Daymon, thank you for your kind words.

It is not correct to infer that the potential pixel resolution would be 481,435 pixels or 23.2% of HD's max resolution.

As you know, the resolution is measured in two axis, the horizontal and vertical, separately. The spatial frequency response measures LW/PH (line widths per picture height). So the first flaw in your logic is that the horizontal resolution (730 line widths per picture height) has to be multiplied by the aspect ratio (16/9) to equate to the horizontal panel size, for say 1298 horizontal pixels.

But even that doesn't tell the story. The MTF50 is the frequency of line widths (more lines = higher frequency) at which the contrast drops to 50% of the low frequency contrast. Put another way, think of the low frequency contrast as an entire screen of solid white or black. But if you break it up into vertical or horizontal stripes that get thinner and closer together, near the limits of resolution, the contrast drops as the whites and blacks move toward gray. When the gray level reaches 50%, you count the lines and that's your resolution.

But actual detail doesn't stop at the 50% contrast level. It's just the point where we stop counting lines. It's arbitrary in that sense, but a consistent method that reflects perceived sharpness.

The extinction limit of resolution goes farther to the point where you can't distinguish fine detail from noise, and it happens at the Nyquist frequency, which you can see on the plots.

I'm curious to know how the comparable cameras would do in this sort of test.

Tom,

Thanks very much for your thorough explanation. Greatly appreciated. It will take a few reads for it to sink in (when i get the time :)). So i'll ask this question simply because i'm sure you'd have the know-how to answer (or say it simply doesnt work like that). :) I'm just wondering what would be the maximum the format *could* get to if we were to ignore things such as cost etc and say Perfection or maximum you could really expect is "this" number? To be able to compare.

The reason i wonder this is because i feel that even though there is beautiful images coming from HD cameras more and more and i'd take them over yesturdays SD stuff - there is still much further they can go. Its a lot of cash to plop down so i'd really like to get more toward the top end then not and perhaps sit out another round with a HV10 to keep me occupied. :)


Tony, i'd love to see that as well. :)

Email me a frame grab of an ISO 12233 chart and I'll run the test for any HDV cam.

You're in luck because I also have the HV10 which I'm going to test tomorrow night, and post the results in the HV10 forum.

I'm already certain it will do well on resolution testing, maybe even the best?

So are these results good?

I think any of the other comparable cameras would fare the same, but unfortunately I don't own any of the other ones to post a res chart or else I would.

Res Charts Anybody?

Tom, this is very interesting stuff. Thanks for the intro and sharing your results. Would you explain why are there more horizontal lines than vertical?

More horizontal lines on the chart, like say 973 instead of what I posted 730? Very simple. The native HDV frame grab is 1440 x 1080 but it is displayed at 1920 x 1080. So the choice is to resize the frame grab, or apply a coefficient of 0.75x (equal to 1440/1920) which is what I did.

Soon I will be posting some interesting results on the HV10, some updated numbers on the XH-A1 that I took at the same time, and finally a definitive answer on that question everyone wants to know, "how much vertical resolution do you lose when shooting 24F."

When I asked before if this was good or not, I guess more of what my very general question was asking was: Do these numbers line up with what canon has quoted for the resolution of this camera as being?

...and or what is the translation of your numbers in x and y pixels?

Jack, none of the camera companies, Canon included, make specific resolution claims on the video files their video cameras will create. The closest to resolution claims I've seen recently is Canon's statement that their new XL 6X lens (note: lens not whole camera) will resolve over 800 lines. They release a subset of specifications, such as pixel count on the sensors, recording format, etc. -- and in fact usually not enough of those to satisfy the more technologically inquisitive customers.

That's probably wise on the part of the companies, given the myriad ways to test cameras, and the "myriad to the myriad power" ways people choose to intrepret those tests. They leave it to reviewers and owners to determine whether the images meet their needs. So there are no rez claims to verify or refute.

For those shopping for a new camera and trying to decide, there are quite a few clips linked here on DVinfo for download and review; more are being posted all the time. Take a look and see if a particular camera's features and image suit your style and needs. If so, then arrange to put one in your hands to make sure you'll like shooting with it. Rez charts and other analyses are useful and interesting -- I do them myself -- but only tell a small part of the whole story of what a camera can do for you and whether it is one you'll be happy owning.

Ditto what Pete said. You don't use these numbers alone to base a camera decision on, and since you're looking for resolution to be stated as a translation of x pixels and y pixels, you don't have enough of an understanding of resolution measurements to be anything other than mislead by them if you did. That's not a knock. Thousands of consumers make purchases every day based on resolution stated as pixels. This HDTV has 1080p resolution, or this one has 1920x1080 resolution. That's just the native panel size Jack. The resolution of detail is always something lower, and it's expressed very differently in any case.

For example, horizontal resolution numbers always seem low. A panel has 1920 horizontal pixels. How come I only get 800 lines horizontal resolution from this cam? For one thing, resolution is most often expressed in line widths per picture height. In other words, the lines that can be resolved are only counted in the horizontal direction for a distance equal to the height of the screen. If you counted 810 of them this way, the number seems low but you would be at the Nyquist limit of the HDV format, i.e. nothing left to be had. Multiply 810 times the aspect ratio 16:9 (1920/1080), and you come up with 1440 lines, which is the Nyquist limit for the format. You may still be able to count more lines but beyond the Nyquist frequency they are just a form of noise or aliasing.

The Nyquist limit for HDV expressed as LW/PH is 810x1080. As lines get closer together, black and white begins to blur toward gray. The MTF50 spec (Modulation Transfer Frequency) says that we stop counting lines when the contrast level gets to 50% (gray). Other measurements of resolution will extend farther, but we need one that is standardized and agreed upon. The MTF50 is pretty universally accepted. But that doesn't stop someone from saying, "hey I see more lines than that just by looking at the chart!" MTF50 most arguments if the testers just agree to use it, which most pro testers do, but I've seen plenty of numbers thrown around where you know they are just eyeballing a chart that's been photographed. A result could be off by 100 lines doing that, not precise enough!

Consider this, you have a new HDTV with 1920x1080 panel size. You project a computer generated image of 960 horizontal black bars, and 480 vertical black bars against white screen backdrop. At best what would it look like? A checkerboard if you're imagninative, or simply a noisy gray picture.

Updated Result XHA1
 

810.0 Lines Horizontal
660.4 Lines Vertical

I've duplicated this result over and over now, nothing further to be gained. Basically it's this, the MTF50 test easily returns Horizontal lines at 825-850, but the Nyquist frequency is 810, therefore the result gets capped at that. The layman's explanation is that the cam easily maintains more than 50% contrast all the way to the limit of the HDV format in the horizontal direction, excellent.

In the Vertical direction, there's more potential in the HDV format than this cam is yielding.

For comparison, the Canon HV10 has the following resolution relsults:

784.5 Lines Horizontal
704.8 Lines Vertical

Nearly an identical performance in the horizontal to the XHA1, and significantly better on the vertical. The actual measured vertical resolution on the HV10 was only 624.0 lines whereas the actual measured on the XHA1 was 665.9, but when the results were normalized to equalize differences from in-cam sharpening, the HV10 got a nice bonus and the XHA1 received a 5 yard penalty.

And what about Canon 24F Mode on the XHA1?

810.0 Lines Horizontal
590.4 Lines Vertical

...so the 24F penalty is a resolution loss of 10.6%


http://vsdrives.com/graphics/resolut...Horizontal.png
http://vsdrives.com/graphics/resolut...1_Vertical.png
http://vsdrives.com/graphics/resolut...Horizontal.png
http://vsdrives.com/graphics/resolut...0_Vertical.png
http://vsdrives.com/graphics/resolut...zontal_24F.png
http://vsdrives.com/graphics/resolut...rtical_24F.png

Thanks for spending the time to do this. Many people are very interested in this, and you have put the issue to rest. Some people have claimed a 50% loss for frame mode, others stating probably 20%-30% was more realistic, but to hear that it is only 10% is very impressive.

Either way, I am impressed everytime I use 24f. I wouldn't care if it was a 50% loss, the picture it creates is a phenominaly crisp, larger than life expression of reality.

So the single chip HV10 has a higher resolution image than the XH-A1? What gives? 3 native 16 x 9 1/2" CCDS and giant 72mm lens are not resolving more information than a single 1.27" CCD behind a dime sized 37mm lens? Is it the CMOS or what?

Tom, Thanks so much for doing the HV10. What an amazing little camera (what a shame we always get a "pick two of the three" style thing with cameras these days). Perhaps i shall get one in leiu of XH-A2 ;)

NB: out of interest i'd love to see a digital camera shoot native 1080p and have this same test done to see what resolution it yeilds. Shouldn't that be the top limit of 810 etc?

Actually, the A1 has 3x1/3 CCDs, and the HV10 is a single CMOS chip.

BTW Tom, thanks for taking the time to do these tests and providing us with the results.

No, in good light with a static test image, it is about the same -- measured at slightly more in one direction, and slightly less in the other direction. As mentioned earlier, there's SO much more to a camera than static horizontal and vertical resolution. They are important things to know, but these cameras provide very different trade-offs for real world use. The HV10 is tiny so it can go anywhere. The XH cameras have much better low light capability and a plethora of ways to control the image and even change the way some of the camera buttons and controls work.

And Tom, let me add my thanks as well for your efforts here.

I posted MTF curves for three lenses for the XL-H1 several months back. They are at http://www.pbase.com/agamid/image/54766171. Note that the axis is in cycles per picture height (the appropriate units) and that this camera turns in about 350 cycles at 50%. If you consider a half cycle to be a line, which most people tend to do even though it is one side of a sinusoid, this camera would be said to have about 700 lines of resolution.

The reason there is response beyond the Nuyquist frequency (405 cycles) is because the picture is upsampled (in the camera) to 1920 horizontal pixels corresponding to 1080 pixels pph for a Nyquist of 540 cycles pph and then sharpened. The plot at http://www.pbase.com/agamid/image/54766057 shows the reconstructed edge from an ISO 12233 chart which clearly shows this sharpening (the under and overshoots). I expect that the explanation is the same for the cameras being discussed in this thread. Put simply, some of the resolution you see isn't really captured by the camera. But it looks real and results in a breath taking picture so who cares?

Another caveat with MTF is that a measurement or two doesn't tell the whole story. It needs to be measured radially and tangentially at several azimuths and cone angles and at different focal lengths and apertures. It is most interesting that Canon has a web site with extensive MTF data on all their lenses - except the ones that go with their prosumer video cameras. Granted most, if not all, of this data is computed rather than measured but still at least it's data.

PS: The F in MTF stands for 'Function' - not frequency.

Actually, it would be particularly interesting to test the resolution with the shutter at 1/25:

http://www.dvinfo.net/conf/showthrea...671#post579671

True 25p outcome? Or will the resolution drop? Would it be substancially?

*EDIT*Link now fixed.

Well said Pete. I love the image from my Sony HC1 (and HDV camera), but there are jobs were I've fallen back to my old XL1s (an SD camera) for better controls to get the shot because resolution isn't everything. With the G1 I'll get the best of both worlds.

Thanks also to Tom. I'm set on getting a G1.

Thanks for the detailed measurements, especially starting to look at frame mode. Any ideas on how to measure the effect of frame mode on vertical resolution in areas of motion? Maybe using shots from a slow horizontal pan across the resolution chart? You'd probably need a true-progressive scan camera to use as a control (show that with the same movement the progressively scanned image doesn't lose vertical resolution, but the frame-mode image loses xx% compared to no movement).

I would expect the frame-mode voodoo to have a harder time maintaining vertical resolution for areas of motion. Of course, if most of the image is static and resolving at almost 600 vertical lines, it'll still make a pretty picture even if your moving subject is at a lower actual resolution. The penalty is probably larger when the camera moves.

-Terence

a little more information about the tests and caveat
 

Warning Will Robinson!

But just to add a few more observations:

1.) Not unlike Canon DSLRs, "L" series lenses won't always outperform a cheaper standard lens at a given selected aperture, say F5.6 or F8.0, but what does seem to be true about them is that the L's give even consistent performance across a range of focal lengths, zoom ranges and uncompromised at the larger low light aperture openings. That is my observation about the 4.5-90L XHA1 lens as well.

2.) I tested the XHA1 at different combinations of zoom range and aperture to find the limits where performance falls off. Of course, I did not try different focal lengths because that requires moving the actual camera to target distance, which in all of the tests was 78 inches. The gain level was 0 db. All frame grabs are "I-frames."

3.) At the 78 inch focal length, I was able to test the zoom range from about 4.5-40mm, or about 0-40% of the zoom range. The performance across this range (9 to 1 zoom!) was very even, with a slight sweet spot at about the 15% mark.

4.) As A.J. Delange points out, a measurement or two doesn't tell the whole story, measuring radially, tangentially, different azimuths and cone angles etc., is needed. I performed just one, the slanted edge test, and only for the center area of image, but I did it as precisely and consistently as I could.

Let's get on with the conclusions!

5.) The range of apertures yielding the generally highest resolution results is F3.2-F4.8, with F4.0 as the sweetspot.

6.) As mentioned earlier, performance is very consistent across the zoom range of 0-40%, with 15% zoom mark being the sweet spot for resolution.

7.) There are all the usual reasons for deviating from these ranges, depth of field control being one, CA generally being less at the smaller aperture openings, (F5.6+).

...and then this observation:

8.) In-Cam sharpening, where do think you have it set? Because from my inconclusive observations, it appears not to be a static amount but changes dynamically with aperture and *gain*, the intent I would infer to maintain the same generally perceived sharpness at low light, high gain, large aperture opening that the viewer would observe in bright light, low gain, small aperture opening. The cam would appear to maintain really sharp detail in low light conditions compared to other HDV cams I've used.

I appreciate the many thanks expressed, but my motivation is that you get something positive and find the information useful for your own purposes.

Tom

One way you could test motion resolution is to use the remote control to zoom in and out. You could do it once as 60i and the second time as 30F and then match up the same exact moment in time and test the resolution.

I'm not sure how the results would turn up because video in motion has lower detail anyways due to motion blur and a lot of other things. Even if resolution of 24F and 30F drops a little bit more during motion does it really matter?

Thank you! I was definitely hoping someone could/would answer that.

I look forward to spending some time reviewing the links you posted to your own tests.

Hey Tom thanks for doing the complicated tests, I know this is off-topic but it would be interesting to run your res test on the RED footage that is available -- just for some sort of wow factor perspective. I for one would like to know what sort of resolution divide we are dealing with between these cams.

You and me both. Point me to a clip of an ISO12233 test chart.

Super job Tom, thanks a lot.

You are quite welcome but it occured to me that the explanation I gave, while factual and a factor, is not the whole story. What is mainly responsible for the high effective sampling rate is that the edge is modeled by some function (I use a sum of Chebychev polynomials) that best fits it and that the derivative of the edge is then calculated from the fit function - not the actual data. Because the analyst computes the derivatives he may do so at points as closely spaced as he choses and thus the effective sampling frequency can be much higher than the pixel pitch in the camera. The MTF which results is the MTF of the model and is not the same as the MTF which would be measured if sinusoidal targets were photographed. It is a good indicator of average image sharpness though because the average transition from light to dark is accurately modeled.

I had suggested panning left-right to avoid having vertical motion blur (assuming a perfectly level camera). Its definitely an academic question. I'm mostly interested in it in wondering how good frame mode is compared to a true progressive frame (the resolution hit is probably greater than 10.6%, but how much greater?).

-Terence

Excellent explanation A.J., thank you.

The only problem is that you would have to do the pan exactly the same each time for it to be fair. You would have to have some automated tripod head that could repeat a pan. The other thing you could do is shoot something that could be made to move past the camera at a constant speed such as a remote control car or train set. This way you should be able to get two frames of the exact same type of motion and the exact same level of motion blur to find out exactly how much softer the F mode really is in motion of the same type of motion. That was the only reason why I suggested the remote control zoom because it would move at a constant speed and would give the same results every time.

Very true. That's probably why no one has ever answered the question of how much vertical resolution is lost with frame mode. If anyone cares to donate a Canon A1 and Sony V1 to me (for a progressive control), I'd be glad to spend a day generating the data....

I was at the local trolley museum on Sunday with my nephew (who is very into trains), and now I have visions of Lionel trains carrying resolution charts in my head....

-Terence

Updated Tests: (01/01/2007) XHA1, 60i and 30F
 

XH-A1 60i: (01/01/2007)
824.3 Lines Horizontal (MTF50); 810 Lines Horizontal = Limit@Nyquist
674.8 Lines Vertical (MTF50)

XH-A1 30F: (01/01/2007)
823.6 Lines Horizontal (MTF50); 810 Lines Horizontal = Limit@Nyquist
592.2 Lines Vertical (MTF50)

Previous Reported Result for XH-A1 24F:
810 Lines Horizontal (Limit@Nyquist)
590.4 Lines Vertical (MTF50)

Previous Reported Result for Canon HV10:
784.5 Lines Horizontal (MTF50)
704.8 Lines Vertical (MTF50)

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

I re-ran the XH-A1 test for 60i and 30F for a baseline. The results are normalized for standard 2-pixel sharpening. Everything is consistent with, in fact nearly identical to the previous run.

The finding that 24F loses a bit of vertical resolution compared to 60i holds true for 30F as well, about 12%, not much and hardly observable if at all.

Limit@Nyquist means the camera has reached the resolution limit of the HDV format in the horizontal direction. 810 x 16/9 aspect = 1440. Remember, horizontal resolution is expressed per picture height.

http://vsdrives.com/graphics/resolut...1_60i_Horz.PNG
http://vsdrives.com/graphics/resolut...1_60i_Vert.PNG
http://vsdrives.com/graphics/resolut...1_30F_Horz.PNG
http://vsdrives.com/graphics/resolut...1_30F_Vert.PNG

Thanks Tom great work. That HV10 still amazes me... what a steller little cam. *still pondering wether to wait for a HV10 mk2 or not*