"The technology explains that the sensor colour filters are actually arranged in a 4x4 subpixel arrangement, meaning chroma resolution is at a native 16MP"
That is a typo. 64 MP with 16 MP chroma resolution means a 2x2 standard Bayer filter arrangement, the same as 99.99% of sensors out there. If you click through to the link, the Samsung press release explains this in more detail.
Looking at Honor's comments about the GM1 and other articles it appears that these are actually 12 and 16MP sensors with a software interpolated higher resolution. Samsung state "Tetracell uses a re-mosaic algorithm to produce full resolution images."
The Honor post suggests otherwise. Honor use both the Sony and Samsung 48MP sensors and they were explaining why they charge more for the Sony version. Samsung's wording also would not make sense if it was a big standard sensor.
source at the very bottom "**Tetracell technology merges four pixels into one so that users can achieve high quality images in all light conditions." (2x2 = 4, 4x4 =16) The press release on Samsung's site also says"***In bright environments, Tetracell uses a re-mosaic algorithm to produce full resolution images."
So Samsung basically shrunk down the pixels to fit more pixels on a phone friendly size chip. In doing so though they killed the low light performance so much that they regroup the pixels in software... Genius! /s... I thought the phone pixels wars ended a few years ago.
What you call "regroup the pixels in software" is how virtually every color image sensor made in the last 30 years works. Don't be confused by Samsung's marketing. This is a conventional image sensor, same as the one in your current phone, just with slightly smaller pixels.
Colour information tends to be lower spatial frequency, so it isn't as bad as it looks. The tiny pixels are more of an issue as noise degrades colour in dark areas very effectively.
Every time some camera chip manufacturer expresses dynamic range in decibels instead of EVs you know they're lying. It's a neat way to cheat by picking a single SNR value in the blacks, where you can easily go beyond 100 dB, while the same sensor performs dramatically worse when measured across the full spectrum.
No optical system with at least one lens in front of the sensor has ever achieved more than 100 dB. The best Nikon DSLR sensors can do about ~85 dB. This toy here probably can't even go beyond the simple lightbox setups which are used to test sensors in the range up to 35 dB.
That number is not SNR, it is dynamic range, which is the ratio of the read noise in the sensor (typically 3 or 4 photons) to the maximum number of photons the sensor can detect without saturation. dB is the standard unit for dynamic range in sensor characterization. Since they're talking about dual conversion gain (essentially, an option to link an extra capacitor to each pixel), dynamic range can be enormous.
100dB for just a sensor is not very high (you can buy lots of sensors higher than that), although it is impressive given the very small pixel size.
useless megapixel race. good thing, the sensor is quite large at 1/1.7. but then, it won't be even able to resolve detail like a 16 MP APSC sensor camera.
Small form factor cameras like this generally have much higher detail than larger format sensors due to the very short back focal length. Some current generation cameras will produce an angular resolution that is not even physically possible on a 16 MP APS-C sensor.
Images from most small cameras like this can be reduced 50% in each dimension without actually losing any detail. If the new sensor is as good, then it'll pick up a commendable 16 MP, but for some reason I'm not getting my hopes up.
This is incorrect. Modern cellphone lenses have MTFs that are close to the diffraction limit (which isn't too surprising, they typically have 6+ aspherical surfaces and a flange distance of < 1 mm, and lens which are designed specifically for the sensor), so if you try to decimate by 2, you'll get both a substantial loss of detail and aliasing.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
17 Comments
Back to Article
peevee - Thursday, May 9, 2019 - link
"The technology explains that the sensor colour filters are actually arranged in a 4x4 subpixel arrangement, meaning chroma resolution is at a native 16MP"So it is useless.
PeachNCream - Thursday, May 9, 2019 - link
Confirmed. Nothing to see here. Move along.saratoga4 - Thursday, May 9, 2019 - link
That is a typo. 64 MP with 16 MP chroma resolution means a 2x2 standard Bayer filter arrangement, the same as 99.99% of sensors out there. If you click through to the link, the Samsung press release explains this in more detail.BedfordTim - Friday, May 10, 2019 - link
Looking at Honor's comments about the GM1 and other articles it appears that these are actually 12 and 16MP sensors with a software interpolated higher resolution. Samsung state "Tetracell uses a re-mosaic algorithm to produce full resolution images."saratoga4 - Friday, May 10, 2019 - link
All image sensors use a 2x2 mosaic which is interpolated up to the full resolution. See: https://en.wikipedia.org/wiki/DebayeringTetracell is no different than anything else.
BedfordTim - Friday, May 10, 2019 - link
The Honor post suggests otherwise. Honor use both the Sony and Samsung 48MP sensors and they were explaining why they charge more for the Sony version. Samsung's wording also would not make sense if it was a big standard sensor.BedfordTim - Friday, May 10, 2019 - link
Edit. bog not big.saratoga4 - Saturday, May 11, 2019 - link
You're just misunderstanding how debayering works.Skeptical123 - Saturday, May 11, 2019 - link
source at the very bottom "**Tetracell technology merges four pixels into one so that users can achieve high quality images in all light conditions." (2x2 = 4, 4x4 =16) The press release on Samsung's site also says"***In bright environments, Tetracell uses a re-mosaic algorithm to produce full resolution images."So Samsung basically shrunk down the pixels to fit more pixels on a phone friendly size chip. In doing so though they killed the low light performance so much that they regroup the pixels in software... Genius! /s... I thought the phone pixels wars ended a few years ago.
saratoga4 - Saturday, May 11, 2019 - link
What you call "regroup the pixels in software" is how virtually every color image sensor made in the last 30 years works. Don't be confused by Samsung's marketing. This is a conventional image sensor, same as the one in your current phone, just with slightly smaller pixels.BedfordTim - Thursday, May 9, 2019 - link
Colour information tends to be lower spatial frequency, so it isn't as bad as it looks. The tiny pixels are more of an issue as noise degrades colour in dark areas very effectively.johannesburgel - Thursday, May 9, 2019 - link
Every time some camera chip manufacturer expresses dynamic range in decibels instead of EVs you know they're lying. It's a neat way to cheat by picking a single SNR value in the blacks, where you can easily go beyond 100 dB, while the same sensor performs dramatically worse when measured across the full spectrum.No optical system with at least one lens in front of the sensor has ever achieved more than 100 dB. The best Nikon DSLR sensors can do about ~85 dB. This toy here probably can't even go beyond the simple lightbox setups which are used to test sensors in the range up to 35 dB.
saratoga4 - Thursday, May 9, 2019 - link
That number is not SNR, it is dynamic range, which is the ratio of the read noise in the sensor (typically 3 or 4 photons) to the maximum number of photons the sensor can detect without saturation. dB is the standard unit for dynamic range in sensor characterization. Since they're talking about dual conversion gain (essentially, an option to link an extra capacitor to each pixel), dynamic range can be enormous.100dB for just a sensor is not very high (you can buy lots of sensors higher than that), although it is impressive given the very small pixel size.
zodiacfml - Friday, May 10, 2019 - link
useless megapixel race. good thing, the sensor is quite large at 1/1.7. but then, it won't be even able to resolve detail like a 16 MP APSC sensor camera.saratoga4 - Friday, May 10, 2019 - link
Small form factor cameras like this generally have much higher detail than larger format sensors due to the very short back focal length. Some current generation cameras will produce an angular resolution that is not even physically possible on a 16 MP APS-C sensor.mkozakewich - Saturday, May 11, 2019 - link
Images from most small cameras like this can be reduced 50% in each dimension without actually losing any detail. If the new sensor is as good, then it'll pick up a commendable 16 MP, but for some reason I'm not getting my hopes up.saratoga4 - Saturday, May 11, 2019 - link
This is incorrect. Modern cellphone lenses have MTFs that are close to the diffraction limit (which isn't too surprising, they typically have 6+ aspherical surfaces and a flange distance of < 1 mm, and lens which are designed specifically for the sensor), so if you try to decimate by 2, you'll get both a substantial loss of detail and aliasing.