Samsung's 108MP camera sensor is getting official on August 12
- ?
- Anonymous
- Ke1
- 10 Aug 2019
sadh, 10 Aug 2019Only PureView does the job. 41MP with 1/1.2 sensor size and... moreMaybe there's a reason why they don't do this again. Given what smartphones need to pack into the chassis nowadays, having the same camera again will make it super thick and it won't sell.
- S
- SteveFox
- ntk
- 10 Aug 2019
Still don't think this will be 27MP. Perhaps 12MP night shots (with 6 in 1 binning for 1.8 microns pixel size) and 24MP, or even 54MP day shots (when binning isn't required).
Still can't understand why the 48MP sensor doesn't do 24MP shots during the day when anything above 1.2 microns isn't required.
- m
- mehtasameer2087
- 7kh
- 10 Aug 2019
Nick Tagataka, 10 Aug 2019I mean, back then the pixel pitch of 20MP sensors for mobil... moreWhat is the surety that all the pixels would light up fully to give you a perfect 1.6 micron pixel data ,it's never going to happen no matter how advanced the tech is now,the pixel binning will never give a fuller potential to what a single large pixel would give.and yes lens quality does matter ,I am sure xaiomi with that cheap pricing doesn't even use glass lenses it might be using fibre glass or some other junk,so no matter they pack 1000 mp into the sensor but the output will still remain cheap Chinese junk paintings no 48mp camera as of now has beaten the proper flagships like smasung note9,iPhone X ,pixel and they still use the standard 12mp sensor ,so pixel size does matter a lot on smartphone or a DSLR,if it didn't then why don't these smartphones having 48mp cameras beat DSLR ,it's the sensor size plus pixel size both
- L
- Let me eiiiiiiiiin
- tDR
- 10 Aug 2019
Let me eiiiiiiiiin, 10 Aug 2019Current AI based demosaic is hit or miss, and requires trai... moreReplying to @Nick Tagataka
- L
- Let me eiiiiiiiiin
- tDR
- 10 Aug 2019
Current AI based demosaic is hit or miss, and requires training (just like all AI in general) and massive computation power, unless they put pre-installed training database. I am not sure but to get 95% hit rate probably requires hundreds MB or maybe GB training data loaded into phone memory and strong AI core to get decent speed. Well reducing hit rate to 80% will reduce memory needed to dozen MB...
That's probably why Huawei need "multi frames", so their AI can pick what they think is the best one, reducing miss rate.
- z
- zakariasotto
- nmi
- 10 Aug 2019
maybe 'turning' 9 pixels into one for added detail, 12 Mpixel
- N
- Nick Tagataka
- vmJ
- 10 Aug 2019
Exyvia, 10 Aug 2019I remember back in the days where people complained 20MP wa... moreI mean, back then the pixel pitch of 20MP sensors for mobile phone was minuscule at only 1 to 1.12µm. Assuming that the 108MP sensor would have the same pixel pitch as Samsung's own GW1, it would have 1.6µm pixels when pixel binned and be 125% larger than IMX586 and GM1 which have already proven to be a very capable sensor in 12MP mode. The dynamic range in that case would remain the same but the low light performance will be enhanced tremendously and so will the detail capture.
Though I must agree with the point you made on the lens quality - I'm not quite sure if Xiaomi is really capable of designing lens that can fully resolve the resolution of 27MP let alone 108. There's a reason why Nokia and Huawei tied up with famous camera companies - it's not just about branding, the partnership can be also helpful when it comes to designing sharp lens as those camera companies have their own set of tests and standards that the phone manufactures need to pass in order to use their names.
- ?
- Anonymous
- 7tp
- 10 Aug 2019
This is not so concise.
They are likely not really designed to only give you the same image as a sensor a quarter the resolution, then why not just use a sensor a quarter the resolution? Using a sensor a quarter of the resolution would likely give you a little advantage due to fill factor (the size used for sensor pixel compared to the surrounding space bordering it) and likely circuitry size advantages that read the value and perform tricks on it to get better pictures. It is easier to get better fill factor and larger better circuitry on larger pixels. :unfortunately the normal 48mp is not as good as I would have hoped (and I'm not saying that it is fine the best it could be at that pixel size). The new Samsung seems the first sensible 0.8 micron in the phone market, as it has better high dynamic range, which has Vern fine with out multiple exposures since at least 1976 at 17 stops or so. At 16-17 stops, done right, images don't look so compromised compared to how we see them as far as latitude goes. They can look more silky. Bring it on! Small sensors are often comprised a lot in this way. If the industry wants to dumb down phone camera video, high compression, moderate very low light ability, and slightly dumb color is enough. But they shouldn't have dumbed down latitude, it goes to the heart of quality in an otherwise poor image.
Now, 48-108mp quad Bayer still makes sense in the following ways. You can still calculate good images at 108mp as each pixel is still surrounded by other colored pixels to approximately calculate the other missing colors at each pixel. This also helps in downscaling resolution to 6k/4k/fullhd. Plus, you can bin each quad Bayer pixel in low light, or to produce a simple image at that resolution along with the image artifacts and quality down grade it produces (seen in phones that use them in this way). According to what I have written in the past, you could also run the read circuits in each pixel of a quad group at different gain or readout, to produce more stops of dynamic range per quad Bayer pixel. This needs to be combined with an low pass filtering technique (spreads the light to neighboring pixels to smooth out aliasing artifacts, which are often differences between pixels caused by the missing color being miscalculated, and the gaps between the pixels. But this smooths the image) such as slightly out of focus technique I promote, which is still in focus at the bigger quad pixel level. However, something in the I image can still pass through the perfect focal distances and produce pixel artifacts there. In reality, what you are focusing on with this technique will potentially be surrounded by an area of these full resolution pixels that may have artifacts. An interesting and acceptable look. However, back onto quad bayering, it means the optical low pass filter (if one is used at all) is likely either adjusted for the 0.8 or the 1.6 micron quad Bayer pixel size. If correctly adjusted for the 0.8 full resolution size, the quarter resolution quadbayer 1.6 micron pixel image will look more artifacted, and if adjusted for 1.6 micro quad Bayer resolution, the full resolution 0.8 micro pixel images will look smoother. I suspect this might be what is happening. However, using a quad Bayer sensor to produce quad Bayer images then recalculate the full resolution images, is going look smooth and not as good as doing it at full in the first place with optical low pass filter adjusted for 0.8 microns.
If you think this is complex, imagine how I would feel watching all this rubbish go on in the phone video camera market?
The phone camera market should go to virtually only virtual color filtering pixels. Where each pixel can detect 3 or more colours, which many major camera companies have been designing test versions of in recent years. It gives superior artifacting and color correctness to Bayer, and you don't need 48mp to get good 12mp images, and produces superior sharpness and resolution compared to 12mp Bayer. The original company that did this did not release much and lacked a lot of other superior technology, but Sony is one of the companies producing prototypes. For video, it is a little tricky video wise, and dies have some issues, but for image compromised phone video camera, it is a good simple fit. The industry could use slightly bad color along with one exposure HDR, moderate very low light performance, and their very high compression, but still produce a reasonable image. It seldom has been used in phones, or in cameras. But going on previous statements, an 4k vertical color filtering sensor will produce an image comparable to a 8k Bayer with optical low pass filtering. An 32mp 8k version, would be as much as you need. 108mp+ unseeded. The thing about Bayer, is it is so poor, they want to move to 128mp+ (192mp+ imager in still mode) to get similar results. But, you can make a 32mp vertical color filtering sensor smaller and cheaper for compromised video function, but people will be fine with a 0.8 micron 4k version, less than 4mm wide. In the now coming computational imaging array (multi point imaging) cameras, 4mm imagers are winners to fit imagers and lens systems in low profile phones. Hopefully the industry gets the drift. Any flagship phone version can presumably do better than current quad Bayer versions.
Thank you.
- ?
- Anonymous
- DkR
- 10 Aug 2019
sadh, 10 Aug 2019Only PureView does the job. 41MP with 1/1.2 sensor size and... moreI did not understand the difference between Nokia's pureview camera & this new quad bayer technology. I thought both use pixel binning so both should produce similar output.
- w
- why
- tEZ
- 10 Aug 2019
why cheap phone using big megapixel sensor & expensive phone just using 12mp sensor ?
& why big megapixel sensor not using "dualpixel pdaf autofocus" for more accurate focus at low light ?
- ?
- Anonymous
- LQu
- 10 Aug 2019
Exyvia, 10 Aug 2019I remember back in the days where people complained 20MP wa... moreBecause most compare photos in wrong way.
Phone A = 12MP
Phone B = 40MP
You have to scale both to same resolution before compared them. The 40MP offers much more crop, it will always look uglier than the 12MP one no matter software processing.
But both scales to 10MP, the 40MP could be the best one in the end.
Also it is wrong to compare using different settings. Different settings will give even more different results.
- ?
- Anonymous
- 8x{
- 10 Aug 2019
So it will take 27 mp images at regular quality
- ?
- Anonymous
- pQr
- 10 Aug 2019
I hope that it is a conventional Bayer sensor. Quad bayer sensors produce too many demosaicing artifacts. It makes more sense to combine a high resolution (conventional) Bayer sensor with an additional low resolution Bayer sensor, but apparently that's too expensive for smartphones manufacturers.
- E
- Exyvia
- YPb
- 10 Aug 2019
I remember back in the days where people complained 20MP was too much. Now were going back up because Marketing.
More MP doesn't always = better.
You'll probably never even see the difference in a phone, I'll prefer larger pixel size and dynamic range. You'll need better lenses as at this level, imperfections of the lens will show.
- ?
- Anonymous
- Ibx
- 10 Aug 2019
Anonymous, 10 Aug 2019What's with all the megapixels that won't actually bring an... morewhy is with the new innovation ? uhmmm
- N
- Nick Tagataka
- vmJ
- 10 Aug 2019
Let me eiiiiiiiiin, 10 Aug 2019Losing more detail. Pixel bining mostly to reduce noise for... moreDemosaicing artefacts can be eliminated by employing good multi-frame processing, Honor View 20's AI Ultra Clarity mode is a prime example.
- N
- Nick Tagataka
- vmJ
- 10 Aug 2019
Blutbad-fuchsbau, 10 Aug 2019108MP camera with 108MB image file sizes. Are you sure this... moreIt would be more like 200MB in RAW.
- Blutbad-fuchsbau
- XTc
- 10 Aug 2019
108MP camera with 108MB image file sizes. Are you sure this cam will be worth it?
- L
- Let me eiiiiiiiiin
- tDR
- 10 Aug 2019
Let me eiiiiiiiiin, 10 Aug 2019Losing more detail. Pixel bining mostly to reduce noise for... moreGood downscaling algorithm using lanczos & spline (or AI based) in real-time probably too heavy for mainstream use SoC. Bicubic and billinear is trash in comparison but used most because very efficient. Meanwhile, Pixel binning offload everything on sensor, both Samsung and Sony got different algorithm and their quality clearly differ even thought their spec mostly same. Huawei also use unique algorithm while using same Samsung sensor