UKC

I haven't used smartphones for photography much, but have tried using an older cheap Huawei and then recently a MotoG31, and both produce quite obvious worm-like artifacts in heavily textured bits of natural scenes -- like rocks and leaves. 

I've never really seen that level of distortion before in images from my other cameras (all mostly just "reasonable" point-and-shoots).

Are the JPEG encoder or sensors in these cheaper smartphone cameras just really bad?

They don't seem very usable.

eg. https://imgur.com/a/4dEpxie  rocks and leaves in bright sunlight

Smartphones have rather small sensors, so they typically are quite aggresive with noise reduction and sharpening. This lloks to be the case here, rather than just jpeg artifacts. You may have some settings to limits the postprocessing. Or try raw if supported? If your native app doesn't support these, then try OpenCamera from Google Play Store. Most likely you'll just find that your phone camera is quite limited.

Hi Marek.  Thanks for your reply. 

Now that you mention it, those squiggles do look a lot like overblown Difference-of-Gaussian/Mexican-Hat sharpening artifacts (when the amount and/or radius is set too large).  I'll play around and see if there are any processing options I can turn off. 

I guess I'm just surprised how bad the defaults are, since the whole world seems to use smartphones rather than dedicated cameras for point and shoot stuff.  I thought they would be at least about as good!

I do use Open Camera app, although that pic was taken with the stock camera as it's a new phone.  And I'm not sure I want to have to use raw for most of the stuff I take.  I think I will just stick with a real camera for anything half important 

Thanks again.

Bruce

That doesn't look like a JPEG artefact to me.   JPEG artefacts tend to manifest and jarring differences between the 16x16 pixel macro blocks; whereas the weirdness in your photo has no such "tiled" structure an organically crosses macroblocks.  It looks like a noise reduction algorithm gone horrifically wrong through being applied with inappropriately aggressive settings.

You might find a 3rd party camera app that lets you record raw or JPEG without such egregious processing.

I'm regularly blown away by the kind of photos my iPhone 11 Max Pro takes, just astounded.  With the "2x zoom" camera and the wide-angle camera it's really freed up my dSLR (and now mirrorless) usage as I know the iPhone will do wider shots just fine, so the discrete camera only carriers long focal length lenses and between them I'm covered.

Cheers.  Yes now that Marek and you have mentioned NR and sharpening, I see lots of references to smartphone images with excessive use of this.  And I agree, that looks like whats going on.

So I'll have a play with settings and see if I can reduce/remove it.  I don't really want to have to fluff around with raw. 

I see the Open Camera app has options for edge processing even in JPEG mode, so might be a way to curb it.

I think Apple have always focused on the arts, haven't they?  So I guess they are going to make their cameras great.

The iPhone camera has a bad a habit of making things look a tad unnatural at times. 

At the end of the day you can't beat physics. Sensor technology (the ultimate limiter in smartphone image quality) doesn't much care about the logo on the phone. Even the iPhone 11 (best smartphone camera?) pixel area is only 5% of a full-frame camera pixel and 20% of a m4/3 pixel (which is already getting a bit 'challenged' under poor lighting).

That's not to say smartphone cameras don't have their uses, but you have to be aware of their (and any camera's) limitations. Just don't expect realistic fine detail from a smartphone or everyday convenience from a FF camera.

Cheap smartphones make up for their poor optics and sensors by aggressively sharpening, as you see here.

Higher end models with better cameras are generally better....and can make some great images when viewed at a macro level, but it's usually quite obvious when you pixel-peek that it's from a phone.

You're right, but current sensors are also nowhere near the limits of Physics in terms of light collection efficiency.  Right off the bat, 2/3rds of all light falling on them is thrown away by the Bayer array.  It's not beyond the wit of man to design out this limit and get more light, almost halving the effect of shot noise on images - shot noise being a major issue on these tiny pixels.

1. A decade ago I had high hopes for Foveon; development still drags on lethargically under Sigma
2. I've lost track of the number of patents I've skimmed with different micro-mirror and micro-prism arrays to avoid throwing most light away by separating colours.  None have come to market AFAIK - I haven't dug in to what the barriers are.

Computational photography is emerging as a powerhouse on mobile devices - fusing data from multiple sensors, correcting motion with the inertial sensor suite on the devices etc.  The computational depth-of-field simulation on the iPhones with the 2D time-of-flight LIDAR system is really very good in about half the settings I've tried it in.  Some of the long exposure hand-held star field images I've taken are staggering in terms of technical capability; the images are nowhere near those from a dSLR in quality but that they can be done at all is just amazing.  

Absolutely, but I've been blown away in the last few years with how far the multi-sensor prime lens approach with computational fusion has come with some mobile devices.   It does make me wonder what the future holds in store for large frame devices as they gain more of the "computational" aspect.  I've recently got the RF 800 mm F/11 prime from Canon - also amazing that this comes in at 10% of the cost of previous long primes and the image stabilisation (lens only, not got an in-body IS camera) is phenomenal.  The lens really needs a wide-field sighting camera integrated onto the outside that talks to the body and feeds a digitally composited EVF display to track fast-moving birds etc.

I'm no longer doing cameras and optics design professionally; but as a hobbyist consumer it's very exciting to see what best-in-class devices are like both small and large, and to think about where it's all going.

Indeed, but it also opens up an enormous bag of worms with respect to 'looks-good' vs. 'is-realistic'. One of my photo-genres interests is astrophotography - I've written my own GPU code to do spatially variant RL deconvolution to try and squeeze more performance in wide-field images, so I know how easy it is to create nice looking fiction!

Ooh, I've wondered about that one! Particularly in comparison with my wife's 400mm F/6.3 on a m4/3 body (so same about field of view). I generally reckon of the difference between m4/3 and FF being about 1.5 stops in term of noise (in my experience based on 6d vs. G9), so on paper it'll be a close run thing. RF is of course a lot newer than 6d, so... Pity about the price of the RF system in general

The M31 only has 1000px on the short edge and the picture has 2000+ so it's been enlarged (possibly more if you zoomed in a little). Hence you're hitting the limits of the bayer array deconvolution methods which will inevitably introduce these wormy looking things, especially if you then sharpen lots. The Nikon D800 got accused of having similar 'orange peel' effects in Lightroom compared to other DSLRs until it was found out that Lightroom was sharpening D800 files more more than other cameras. 

youtube.com/watch?v=o5t5QmMAqVY&

Can of worms indeed; the "AI superresolution" stuff and a deconvolution done without a conscience both stray in to fictive territory - but then there is a whole range of reasons to take photographs from those where fiction is not tolerable (crime scene photography, medical imaging in dermatology) to those where it's about the artistic merit of the output.  Interesting point - I wonder what the guidance is to police and medical authorities on this area...  When I stop and think about it, it bemuses me that there's 16 teraflop processing engine in my pocket that's used to run ANNs to make my photos look better.  

But many of the computational photography things are most definitely not fiction, rather an attempt to capture as authentic an image as possible through the use of multiple sensors and a priori knowledge of the transfer functions of those sensors and contemporary knowledge of the internal sensor suite.

I've wondered before about an daft extension of this sort of thing - you go to take a photo, the camera device geolocates itself and taps in to the global corpus of photos of that area, and merges those so you can turn your day-time summer photo in to a night-time one or a winter one, or one with real zoom detail your device could never have caught.  Imagine being able to walk around and see back-in-time in real time.  A bit like the device Evil-Lyn used in 1987's Masters of the Universe...

It's just mind blowing.  As you say, the whole system is currently expensive.  Big step up from my 6d.  I don't have an RF camera, borrowing my partners.  When I can find the transfer cable I'll post a hand-held moon shot taken with it.  Finding and tracking moving objects is taking some practice; I've printed a hot-shoe to NATO accessory rail converter and have a holographic site on the camera, but I'm gradually getting there by using two eyes instead, one down the lens and one not.  This gets some fantastic effects from the brain's processing - I see the aiming reticule from the EVF, containing the bird/plane, floating in the right place over the unzoomed sky from my other eye.  

Cheers Tim.

Yes, I think the guts of it seems to be that the sensor size of these smartphones is tiny.  Then they over-sharpen these things to death.

I did some initial quick tests looking at leaves and gravel out my window using Open Camera App.  With Open Camera when you switch to Cam API2 you can control the amount of sharpening.

When you switch sharpening Off, the results are much more pleasant to the eye.  It's more noise and blur rather than bright worms.

As I said above, I'm just surprised how much these phones are used as cameras, given how bad they seem without putting in a bit of effort.  Obviously most people don't care.

For me, point-and-shoot digital cameras from around 20 years ago produced more pleasing results -- for much less pixels/file size.  So I'll still carry a small camera when I want to capture something.

Thanks to everyone for their input.

It's just horses for courses. I use my camera phone quite lot - for snaps to send via WhatApps that never leave the phone ecosystem, or for 'memos'. Ideal for that: always (usually) with me, convenient and adequate quality. Other 'better' cameras come out (more rarely) for wildlife, astro, landscapes and so on. They all have their limitations and uses. 'Better' just depends on requirements.

Yeah, that all makes sense.  I was probably expecting too much. 

I don't take many pics at all, and have gone through years of my life (and some of the best adventures! ) without taking a camera. 

Now that I need a smartphone to occasionally do some basic mobile computing while traveling, I thought might as well use it for the odd pic as well.  Only to find I don't like the results.

Although getting rid of that over sharpening helps heaps.  So this has been a great thread - thanks.

You do have to consider that some phones have better cameras than others. None are optically and sensor-wise as good as a dedicated camera, but the variance between them is huge. The Moto G31 has an 1/2.76" sensor. The iPhone 11 Pro has an 1/2.25" and the 13 Pro 1/1.9". There is a 44% difference there.