This post was shared with us by Werner Ruotsalainen. There was quite a bit of discussion over the post below regarding the image quality of the Nokia Lumia 1020:
It’s a very thorough discussion despite the limited availability of early Nokia Lumia 1020 samples. So you can imagine what Werner will deliver when the device becomes available.
Nokia 1020 Still Image Quality – Any Good?
Currently, I have no access to a production Nokia 1020 yet. (While Nokia’s homeland is Finland, it’ll only appear in shops here only some months later.) This means it’s no possible for me to run my own tests on the handset or compare it directly to, say, the iPhone 5. This, unfortunately, also means only a smallish subset of image quality parameters can be properly examined. In the following, I provide you with an extended version of my original comments HERE.
1.1 The Original PR Images
Nokia has made available several images online on their official blog HERE. Unfortunately, most? all? of these shots, presumably shot on a pre-production device, are certainly inferior to ones shot on production ones – for example, those made by the GSM Arena folks or the main subject of this article, that is, the shots made by an AT&T employee.
(Incidentally, the GSM Arena folks have also emphasized HERE the following: “We are also pleased to report that the blurring along the right side of the frame, which is visible on the samples wasn’t present on our unit so it is definitely due to a imperfect prototype unit.”)
As has been explained, the official Nokia shots don’t seem to represent the image quality of the final version of the handset – they may have been engineering samples with more (or straight non-existing) tolerances. It’s, nevertheless, pretty strange Nokia didn’t strive for publishing less flawed images. This might have been a oversight on their part. A pretty bad one, I’d say – after all, a lot of people in every forum used these images (not having access to anything else) to form a pretty bad conclusion on the image quality of the handset. That is, Nokia did make a major mistake by letting these images out. (We’ve been discussing the why’s of publishing these flawed samples. See for example my discussion with “Janne” HERE (first, dedicated comment at July 17, 2013 at 7:52 pm GMT: “I still wonder why Nokia published those awful (noisy and decentered) demo shotsâ€¦“))
Let me show you some examples of why the official samples are inherently flawed. We’ll concentrate on the most prevalent and software-unfixable issue, major blurring because of decentered lens.
1.1.1 What’s Lens Decentering?
Lens decentering is a major problem well known for every serious photographer. Many known camera (or lens) models suffered from some degree of decentered lens, at least in the first batches; for example, the Canon S100. For example, DPReview have reported (link) the following on the S100′s they’ve received for testing: “…the camera [s100] sitting on my desk in front of me is the third sample that I have looked at, and the third with what appears to be a slightly decentered lens.” The results of the decentering was (also) almost half-framesize blurring on the right. (Fortunately, later S100′s have been fixed and I don’t know of they having major decentering issues.)
The same effects are clearly visible in Nokia’s official samples. In the following two subsections, I elaborate on both 16:9 and 4:3 samples.
22.214.171.124 A 4:3 Sample
On the 4:3 sample I’ve evaluated, about 1850-1900 pixels on the right of the 4:3 shot has major blur problems. It’s not a simple focus point issue – objects in the image’s center and right, on the same focal distance, are perfectly focused and have no blur.
Note that the background buildings in the same (original) image also exhibit some visible oversharpening and detail-smudging issues. The following crop, taken from the upper center of the original image (that is, from an area where the blurring was not an issue any more, unlike the rightmost part of it) shows pretty ugly oversharpening halos around the contour of the building:
In other shots, I haven’t noticed this kind of oversharpening. While the shots are indeed sharpened by default, halos don’t generally become distracting.
126.96.36.199 A 16:9 Sample
The 16:9 sample image HERE exhibits a “dead” zone on the far right of about 1000 pixels:
Note that rest of the images (for example, THIS, THIS and THIS) would have been much harder to correctly evaluate: all images only contain in-focus subjects in either the center-left or the center of the shot.
Fortunately, production units seem to have much smaller areas with (strong) blurring.
2. Edge Softness in the AT&T Images
While the AT&T shots, decentering / blurring-wise, are much better than Nokia’s own, official samples, they still exhibit quite strong edge softness along both the left and right edge. This is pretty much visible even in low-resolution, oversampled images.
On 16:9 shots, the heavily blurred zone is about 750 pixels on both sides – that is, 20% of the entire frame’s width (7712px).
2.1 Fixing the Softness
Edge softness in no way can be fixed. Oversharpening would make things even worse as there’s almost no true source detail to sharpened. In addition, it’s exactly in the soft areas that Chromatic Aberration (CA) is visible (unlike in non-edge areas), further degrading image quality.
All in all, as long as you do require a tack sharp image along the entire picture and/or CA is far too evident and can’t be fixed with the traditional CA fixing tools, your only way of doing this is just cropping – that is, removing the leftmost and rightmost pixels of around 750 pixels in 16:9 and about 750-288 = 462 pixels in 4:3 images.
(How I have got the number 288? The Nokia 1020 is (as with the 808) truly multi-aspect, just like many Panasonic cameras (ZS3/TZ7, LX5, GH1/GH2 etc.). This means 16:9 mode make use more horizontal pixels of the sensor than 4:3 mode, which also means the former mode also makes use of the extreme edges of the lens, unlike the latter. You may want to consult section “Multi-Aspect Sensor” at http://www.dpreview.com/reviews/PanasonicDMCGH2 for more info on multi-aspect sensors. Now, given that 4:3 pics are of resolution 7136×5360, while 16:9 ones are 7712×4352, the left/rightmost (7712-7136)/2=288 pixels are abandoned in 4:3 mode. This also means the blurred zone should be around 750-288 = 462 pixels on the left/right edges of 4:3 shots.)
2.1.1 Consequences of Manual Cropping - Resulting (Equivalent) Focal Length
If you crop from 16:9 images (again, these suffer from lens softness more than 4:3 ones, the latter having considerably thinner blurred left/rightmost image areas), the effective focal length does increase. By simply cropping the 2*10% of 16:9 frame (that is, around 770 pixels on both sides), we’re still at 26 + 2*2.6 = 31.2 mm equiv focal length – with, now, excellent sharpness. It’s still considerably wider than the camera of most other flagships, particularly in video mode. (iPhone 5: 33mm in 4:3 stills / 42 mm in 1080p video – particularly the latter is much-much narrower than the 1020, even with some heavy cropping to get rid of the blurry edges).
188.8.131.52 What About 3:2?
Cropping the 16:9 originals has, in addition to completely getting rid of the blurred left/rightmost image areas, another advantage: getting closer to the 3:2 aspect ratio (to my knowledge) missing from the Camera Pro client of the camera. Assuming you don’t crop vertically, that is. (You don’t need to as, the lens’ being round, there’s no blurring in the shots around the bottom / top edges. Also see my post at July 18, 2013 at 4:51 pm GMT HERE for more info.) As you may know, 3:2 is the aspect ratio used by most large sensors (except for the 4:3 micro 4/3 and the now-discontinued 4/3 sensors.)
Should you remove exactly 2*10% of the frame, that is, 771 pixels, 7712-771*2=6170 pixels remain. If you don’t crop vertically, this will be, at 1.417 aspect ratio, much closer to 3:2 (=1.5) than either the source 16:9 (= 1.777) – or, for that matter, 4:3 (= 1.333).
By cropping somewhat less pixels (or, alternatively, in addition to cropping 771*2 horizontally, also cropping (4352-6170/1.5)/2 = 119 pixels from both the bottom and top), you can achieve precisely the 3:2 aspect ratio.
184.108.40.206 Don’t Forget The Cropping Areas When Shooting!
Of course, cropping as explained above requires you to pay attention to framing your subject. Don’t use the outermost 10% next to the edges! Unfortunately, currently, it seems you can’t use custom rulers in the Camera Pro app and, according to Nokia, accessing the 1020 camera from third-party app (one where you could display 10% rulers to show the area that will need to be cropped) doesn’t allow you to use the built-in, in-camera oversampling method of Nokia. That is, for the time being at least, you’ll need to shoot without cropping area-specific rulers.
3. Other Image Issues
I’ve dedicated the entire second section to blurring resulting from the lens being “pushed” far too hard on the left/right edges. Fortunately, that seems to be the single most important issue with the 1020′s camera; this is why I discuss other issues in the same (this) main section.
3.1 Rolling Shutter Effects in Stills
HERE, I’ve posted quite a lot on the differences of mechanical and electronic shutters and, in general, the inherent problems one faces when using a sensor without a global shutter.
The 1020, as with the 808, doesn’t have a global shutter either. This means there’ll be some image skewing even in still mode where the mechanical shutter can be utilized. (Note that I, not having a reliable video recording of, say, fast-moving cars, I could’t properly evaluate the rolling shutter problems of shots made with purely electronic shutter. I assume it’s somewhat – but not much! – worse than the image skewing distortion visible when using the mechanical shutter.)
It’s actually not that bad, given that we’re speaking of a 41 Mpixel sensor, meaning really a LOT sensor photodiodes to be read. In addition, the car in question had the speed of around 80 km/h (50 mph); that is, it was pretty fast (at least in the eye of somebody used to the traffic speed regulations here in Finland ). I’ve certainly seen much worse results from cameras with much fewer sensor pixels.
3.2 Chromatic Aberration
There’s no Chromatic Aberration (CA) to speak of in the (horizontally) center 80% (that is, the non-blurred area) of the frame, “only” the 20%, where – as I’ve previously mentioned – the lens are pushed too hard, over their limits. CA in the left 10% is very visible for example HERE and HERE:
I used red rectangles in both crops to emphasize the areas with quite strong CA.
Again, if you do crop (as I’ve recommended), you’ll unlikely to be faced by any kind of CA. The lens of the 1020 is VERY well done in this respect (too).
3.3 Dynamic Range (DR)
DR is, as was easy to predict, doesn’t come close to for example APS-C sensors; see for example the abrupt clipping in the green foliage at the bottom of THIS image:
Here, I used red rectangles in the crop to emphasize some (but definitely not all!) of the clipped areas.
However, it seems to be definitely better than that of the 808 and on the same level as better small-sensor P&S cameras.