maemo.org - Talk - Fcam apps VS standard Camera app VS SONY DSC-H9 shoot out! WARNING: VERY LARGE IMAGES

Well, it took two and a half hours, but here we go:

N900 (Fcamera) versus DSLR (Alpha 200/Carl Zeiss 16-80mm@f/4 35 mm)

At first, I took the test with all ISO settings, used LowLight, and as I progressed I realized that some of the limitations kicked the heck out of the N900.

First, it's impossible to mount the N900 to shoot down (as opposed to a good tripod), so I had to re-do the DSLR tests.

Second, "focus" is a joke on the N900. Because of such a limited close usage and small optics, it usually focuses alone the lines of "close" "far" and "infinity". Because of this, and lack of 200-400% view + manual, I was getting results so mixed up I had to move the board again and go lucky imaging on it's silly behind.

Third, I'm fairly sure the accelerometer isn't 3 degrees off, it's the case. Settled on the same surface as the DSLR, the image was slightly rotated. I had to put something under the side of the phone to make it shoot straight.

Fourth, the force needed to hit the shutter is huge compared to the size and weight of the phone. It's virtually impossible to hit the shutter without tilting it, even when standing on a hard surface.

All images are 1280x480, made of two 640x480 images. Because of sensor form factor, A200 will have white bands over and under. In the top left of each image you will see the center of the shot, scaled to 200%.

N900's shots were from the DNG, because they looked better than the JPEGs. Corrected for idiotic white balance, no further processing.

And now, ahead we go.

ISO 100

http://lh4.ggpht.com/_FCBr8WhLe3I/TF.../ISO%20100.jpg

Right off the bat, I ran into trouble. At ISO 100, the N900 was struggling with exposure time. While the DSLR claimed it can hit it in 1/3@f/4, N900 seemed pegged at 1 second, with a lens that has f/2.8 written on it. Not on this sensor and setting.

The quality of the shot is, as you can see, rather horrible. At 1 second exposure and with the hard to push button movement is a fact of life. I have a steady hand, shoot quite a lot and had a hard surface.

Also, it seems that there's some post-processing involved. Don't really know why and by who, but it certainly looks like there's SOME attempt at sharpening.

ISO 400

http://lh3.ggpht.com/_FCBr8WhLe3I/TF.../ISO%20400.jpg

Ah, the more human. At ISO 400, the N900 was still struggling with exposure time. True to its claims, the DSLR claimed it can hit it in 1/13@f/4, N900 still seemed pegged at 1 second or close to it.

I decided to skip ISO 200, there was little difference. Along this comparison, I'll skip a few settings and light levels, as they are not required.

Again with the post processing. Image is horrible, at first I thought i moved the camera, but there's no streak to show which way. There simply isn't enough detail in there. The only explanation I could think of is focus software failure.

No detail, incredibly soft. On the good side, soft edges took the noise out of the picture. If this was a sunset, it would look pretty good, warm tones, no fringes. On a test board, however, a disaster.

ISO 1600

http://lh6.ggpht.com/_FCBr8WhLe3I/TF...ISO%201600.jpg

Now we have something. At ISO 1600, Alpha starts to visibly lose ground. Normally this is outside it's best range (never shoot over 400 for crystal pictures).

N900, however, has newfound life. With the exposure time now lowered enough to be decent, the viewport no longer drags and the darn thing actually focuses. With 1+ second at 100, N900 now does a decent 1/12 or so exposure time at 1600.

Unfortunately, by now the noise is quite high, lots of color noise (hello, CMOS) but we actually see lines on the soft side. It ain't pretty, but we have detail, folks, which means that post-processing is an option now.

ISO 3200

http://lh5.ggpht.com/_FCBr8WhLe3I/TF...ISO%203200.jpg

By now, the DSLR is WAY out of its element. Exposure is way high, started out at 1/3 at ISO 100, by now it shoots at under 1/100, almost 10 times faster than it needs to be. At 1/100, I could take pictures while jumping up and down.

N900, however, is getting started. Noise is through the roof, color noise all around and by now the details in the fine print is killed by noise. There is nothing ahead for us.

=======================================

All above photos were taken indoor, in a 7 by 5 room, lit by about 200W incandescent equivalent. The board was standing up, not in direct light.

Light level is what is normally deemed low, but it is enough to read fine print.

I used light meter app, selected Sony Alpha 200 and Zeiss 16-80 mm, at F/4. It suggested ~1 second (0.8-0.6) exposure.

========================================

I have then decided that to get nice photos, one must limit oneself to ISO 400. By 800, both cameras were getting some noise, and by 1600 it was downright visible.

So I left both cameras at ISO 400, and started lowering the light:

From 200W to

Single Bulb (40W)

http://lh6.ggpht.com/_FCBr8WhLe3I/TF...44/VeryLow.jpg

At this point, the DSLR was quite happy. At ISO 400, it suggested 1/4 from 1/13, which seems to suggest about one third the light. Given that the 40W bulb was closer to the board than the rest, this wasn't a shock. There's some blurring involved, as 1/4 (250 ms), noise reduction starts to kick in. Also, to make it fair, I let IS on, even though I was using a tripod.

The N900, however, is running for the hills. Exposure was way, way up, and even so, the shot was bad. No shaking, the image is pretty clear, but way too dark, and even with post processing there's simply too much noise.

The JPEG isn't much better off, some light is visible, but no luck.

At this point, I have decided to try and kill the DSLR. I saw what N900 can do, let's see what it takes to make the CCD fail.

Zippo (one candela)

http://lh4.ggpht.com/_FCBr8WhLe3I/TF...s144/Zippo.jpg

I was sure I'd break it. I turned off all light sources everywhere, killed everything, even covered the HDD LED from the PC.

I lit my Zippo and held it near the camera, approx 50 cm away. I had trouble focusing, so I set it to manual and took the shot. At ISO 400, it took 3.2 seconds.

If you look at the center of the shot, you will see a bright line. That's because the scotch tape I used to hold the papers together reflected the fire. Also, it's visible by now that the right side of the board is brighter than the left, indicating just how little light there was.

Well, it didn't break.

In the N900 shot, it was too dark so I used another lower-left sample with brightness way, way up. Turns out, it still sees the pattern on the left side. In theory, it could still take some part of the shot.

However, since the DSLR took 3 seconds and is about 3-4 times faster than the N900, I'd guess 9-14 seconds exposure which is wholly and undeniably un-doable.

The 1 second software limitation sends the N900 plummeting when the light fades, but frankly, as a light, poorly held, strong shutter device there is no point in going at such low shutter speeds - there will be no image.

So there we have it.

Conclusions

Let's see.

One, it's not a camera. Not by a long shot, even if you have an atomic shot launcher.

Two, up the ante with the ISO. In low light, higher ISO actually makes the picture usable, even with noise. ISO 400 is not bad on N900. In low light, ISO 400 is basically a requirement.

Three, keep exposure down. Up the ante on the ISO if you must, it's better to have grainy detail than soft nothing. You can soften it up afterwards if that's your thing.

Four, focusing stinks to high heaven, which is understandable from such a camera. I always suspected focusing is a weak point, but not by how much. Remember, every focus point you add throws focus by a lot more as the distance grows. So while focus is an issue at 30-40 cm, by 1 m it's just about simple, and beyond that it just gives up with "infinity" (as it should). So, take your pictures from further away. Better you have a smaller subject than a blurry subject.

Five, remember, because of the button, all shots WILL be in motion, even by a little. So, don't go for slow shots. When ISO is already high (over 1600) and time goes over 1/10 or so, switch to flash. There is little point in shooting dark.

Six, it's about 3 to 4 times slower than the DSLR, which shoots at f/4. The DNG says 1/10 at f/2.8, while the DSLR shoots at 1/25 at f/4. This is a bit confusing for me. Either the values in software are off, or the lens actually focuses on an area much bigger that the sensor. At any rate, the N900 behaves almost full two stops below the DSLR, placing it at f/5-f/6, with the back of the device saying 2.8/5.2. The camera happily signs the images with f/2.8. I'll guess it's actually an f/5.2 optic.

In low light, ante the ISO up. Try to keep exposure under 1/10 if at all possible. Grain at ISO 3200 is WELL below the loss you get by moving the camera since it has no image stabilization.

And finally, it's a small sensor, a CMOS, with limited optics. Try to not take pictures in the dark. Go closer and use the flash. And really, don't hope for 5 MPx.

In fact, I was so curious about detail in low(er) light that I took the shot and went for Nyquist frequency test.

Here we go:

Nyquist Frequency

http://lh3.ggpht.com/_FCBr8WhLe3I/TF...44/Nyquist.jpg

All right.

Looks bad, doesn't it? Because it is.

Shots are taken at ISO 400.

Vertical frequency on N900 goes (loosely) to about 4.2, no more. Above that, the pixels between the lines no longer hit white. In all clarity, about 3.4 before it starts to fade into one another. I use the term "white" loosely, there is no white on N900's shot.

Diagonal is also bad. We have a decent 3, but 4 has lost it. We can guess there's a diagonal there in 6, but it's basically artifacting.

DSLR, however, is doing well. The JPEG here was actually reduced to fit, but the original RAW file actually has the white pixels right up to 10, with slight Moire at 10 (diagonal).

Vertical also goes well, white pixels all the way up to 10. At 9.5 or so the black lines start to go wider and narrower, as the resolution struggles to keep up.

The chart is 36 cm wide by 20cm high (white area, where arrows point).

There are ~40 lines (37) in 1.4 cm, that's 2.85714 lines per millimeter. Given that in order to have fat and thin lines you need 1.5 pixel rounding, that's ~3 pixels per line. Every two lines, one is wider (white). We have 360 mm * 2.85 = 1026 lines, with a minimum of 3.5 pixels per line that's a rough 3600 pixels needed to get that result. Given roundings, and the fact that I get some black on the frame, it's safe to say there's info there for at least 3600 pixels out of the 3800 it puts out. That's 10 MPx of information.

Let's look at N900. Beyond 4.5, I can't even count the lines. It shouldn't be this way, but I'll give you a discount. 4-5 that's 18 lines. That's great, because it's exactly half what 9-10 has, meaning it simplifies calculations.

In stead of 2.85714, we get 1.285714, half-ish. It goes one-and-a-half pixel at about 4-4.2, and goes under-pixel (loses all white) at about 4.5. That's basically 2 pixels (one black, one white, ignore the fade in) per line. Using the same approximation as above, 360 * 1.285714 = 462 lines, that's 925 pixels actual information. At 1.31 ratio, that's 706x925, a grand total of 650KPx.

Yes folks, under low light and at ISO 400, your grand total of information is 650K, not much above 800x600 (about 1.5x), meaning that it will look great on that N900 screen, even be up to a 2x zoom before it starts to show its ugly, ugly face.

At ISO 100 it also flops, about 900K, which is spot on 1280x720. It also means it can take over 2x zoom before it shows its ugly head.

I did the math so many times my head hurts.

Decided to test: I even the images on my phone, measured how much half the image is, and zoomed up to be sure I zoomed 200% (hit the rocker, slide, until it shows 1/2 of the image). To me, that's 5 presses of the rocker (one for wide-fit, 4 for zoom).

By the time it hit 200%, all my night shots showed artifacting. By 300%, I could see blotches and I could see pixels on edges. So, tried and tested, less than 300% zoom. About 300% max, even though with a keen eye one spots stuff at about 200%.

At 800x480 screen, that's 380KPx, at 200% that's 1.5 MPx. 300%, that's 3.5 MPx. For this, however, you need good eyes. And with good eyes, from 200% you no longer get any details.

That's it then. Under tested conditions, standard camera, long ISO 100 exposure, N900 gets under a megapixel.

All is not lost, however. Remember that ISO 800 shot where everything went clear?

Imported into PC, filtered down color noise, aggressively reduced colors to compensate for noise, lost color balance in the process (image looks a bit washed out) and the results are in.

I'll skip the math: N900 delivers a cool 4.6 MPx at the expense of some noise and some color, plus the downside of having to carry the images over for post-processing.

That's it folks. Tune in tomorrow, for sunny tests, full light and color testing and the final word on pixels and ISO.

Good night.