RYYB sensor on Huawei P30 Pro

[Photo_Genius]

Huawei relased the P30 Pro. It has a "Sony-made 40MP Quad Bayer sensor with an RYYB filter" according to this site . Huawei claims that the RYYB sensor collects 40% more light.


DP Review article

Why does a RYYB sensor collects more light than a RGGB sensor? Is that even true?

Are we going to see RYYB sensor in new cameras?

How is it possible to generate an image with red, yellow and green? Those are not primary colors actually. With RGB you can get yellow if you mix green and blue, but with RYB, how do you mix the color green?

 

[J A C S]

Well, a B&W sensor collects even more light...

 

[bizi clop]

Huawei relased the P30 Pro. It has a "Sony-made 40MP Quad Bayer sensor with an RYYB filter" according to this site . Huawei claims that the RYYB sensor collects 40% more light.

DP Review article

Why does a RYYB sensor collects more light than a RGGB sensor? Is that even true?

Are we going to see RYYB sensor in new cameras?

How is it possible to generate an image with red, yellow and green? Those are not primary colors actually. With RGB you can get yellow if you mix green and blue, but with RYB, how do you mix the color green?

RGB .... RGB .... RGB .... RGB .... Incoming white light, 3x4=12 "units" of light
R__ ..... _G_ ..... _G_ ..... __B ..... Bayer Color Filter: 4 units of light
R__ ..... RG_ ..... RG_ .... __B ..... RYYB Color Filter: 6 units of light, 3x more collected red light, 6/4=150%

Green = Yellow - Red

Dpreview is not ascii-graphics friendly.

 

[J A C S]

Huawei relased the P30 Pro. It has a "Sony-made 40MP Quad Bayer sensor with an RYYB filter" according to this site . Huawei claims that the RYYB sensor collects 40% more light.

DP Review article

Why does a RYYB sensor collects more light than a RGGB sensor? Is that even true?

Are we going to see RYYB sensor in new cameras?

How is it possible to generate an image with red, yellow and green? Those are not primary colors actually. With RGB you can get yellow if you mix green and blue, but with RYB, how do you mix the color green?

RGB .... RGB .... RGB .... RGB .... Incoming white light, 3x4=12 "units" of light
R__ ..... _G_ ..... _G_ ..... __B ..... Bayer Color Filter: 4 units of light
R__ ..... RG_ ..... RG_ .... __B ..... RYYB Color Filter: 6 units of light, 3x more collected red light, 6/4=150%

Green = Yellow - Red

Dpreview is not ascii-graphics friendly.

It is not that simple. It is not clear what that Y filter really does. It may not be (it' better not be) R+G. Even if it is, you collect more light but with the color transformation, you get more noise just from that step. The B&W example I gave above is taking that argument to the extreme.

 

[Sympa]

The Y indeed seems to be yellow. "ISO can go up to 409.600" no less.

My guess: it is better in low light if one doesn't care much about color. Perhaps color can be inserted via "AI" methods.

And it is new and different, it is hard to differentiate in the market, and is a plausible story.

There is also a wide angle camera that might be used to fill in 'missing color'...

 

[noisephotographer]

At least when I look at sample images, the RYB sensor doesn't seem to be a noticeable advantage. See https://www.gsmarena.com/huawei_p30_pro_camera_comparison-review-1912p4.php , for example see Night scene 2. The Mate 20 Pro has an f/1.8 1/25s exposure, the P30 Pro an f/1.6 1/33s exposure, so it's nearly exactly the same exposure (1.8/1.6)² x 25/33 ≈0.96. If the RYB sensor leads to 40% more captured light, then the P30 Pro should have used less noise reduction, but the P30 Pro sample image doesn't really look better than the Mate 20 Pro. Of course 0.96x1.4= only 34% more light wouldn't be a very noticeable advantage anyway.

Also see the artifacts section of Dxomark's review. https://www.dxomark.com/huawei-p30-pro-camera-review/ Recently I tried the P30 Pro in a store and it produced very noticeable pink, green blots next to light sources, it definitely didn't look like chromatic aberration.

 

[robert1955]

Huawei relased the P30 Pro. It has a "Sony-made 40MP Quad Bayer sensor with an RYYB filter" according to this site . Huawei claims that the RYYB sensor collects 40% more light.

DP Review article

Why does a RYYB sensor collects more light than a RGGB sensor? Is that even true?

Are we going to see RYYB sensor in new cameras?

How is it possible to generate an image with red, yellow and green? Those are not primary colors actually. With RGB you can get yellow if you mix green and blue, but with RYB, how do you mix the color green?

maybe I'm missing something, but doesn't B stand for Blue?

(the change is replace GG with YY)

 

[John Sheehy]

Green = Yellow - Red

However, green_channel_noise = yellow_channel_noise *PLUS* red_channel_noise, for non-fixed chromatic noise. "Plus" in quadrature, of course.

Benefits would mostly be found in B&W images or the luminance abstraction of a color image.

--
John
http://www.pbase.com/image/55384958.jpg

 

[Jack Hogan]

Photo_Genius wrote: ... but with RYB, how do you mix the color green?

As has been mentioned it depends on the Spectral Sensitivity Function of what they call 'yellow'. Here they show yellow as having a response similar to some portions of green + red. So green = yellow - red. But notably they still have green pixels.



In general, take the green corner of the chromaticity diagram and move it around a bit. Until we see the SSF of 'yellow' I am not sure what we can say about implications on color gamut, color discrimination and noise.

Jack

 

[Entropy512]

Huawei relased the P30 Pro. It has a "Sony-made 40MP Quad Bayer sensor with an RYYB filter" according to this site . Huawei claims that the RYYB sensor collects 40% more light.

DP Review article

Why does a RYYB sensor collects more light than a RGGB sensor? Is that even true?

Are we going to see RYYB sensor in new cameras?

How is it possible to generate an image with red, yellow and green? Those are not primary colors actually. With RGB you can get yellow if you mix green and blue, but with RYB, how do you mix the color green?

RGB .... RGB .... RGB .... RGB .... Incoming white light, 3x4=12 "units" of light
R__ ..... _G_ ..... _G_ ..... __B ..... Bayer Color Filter: 4 units of light
R__ ..... RG_ ..... RG_ .... __B ..... RYYB Color Filter: 6 units of light, 3x more collected red light, 6/4=150%

Green = Yellow - Red

Dpreview is not ascii-graphics friendly.

It is not that simple. It is not clear what that Y filter really does. It may not be (it' better not be) R+G. Even if it is, you collect more light but with the color transformation, you get more noise just from that step. The B&W example I gave above is taking that argument to the extreme.

Or, in short - why every time someone tries to replace Bayer RGB, it usually doesn't last long in the market.

(Fuji X-Trans being the longest-lived of such approaches, although that's still RGB, just a pattern other than Bayer)

 

[Eric Fossum]

Photo_Genius wrote: ... but with RYB, how do you mix the color green?

... So green = yellow - red. But notably they still have green pixels.

still have green pixels? RYYB - no green. Maybe you meant something else?

========

Anyway, in general, it could just be a different color correction matrix with a different set of coefficients. [RGB] = [CCM] x [RYB] where [RYB] is the interpolated color for a particular pixel.

Perhaps [CCM] =

1 0 0

-1 1 0

0 0 1

Well, at least that is the simple approach. Hopefully none of the coefficients are -1 in the CCM. When you try to match a Macbeth chart with the least error (& some chart colors weighted more than others) you will get different coefficients.

(Jack, this CCM part was for the beginner-readers)

 

[alanr0]

Photo_Genius wrote: ... but with RYB, how do you mix the color green?

... So green = yellow - red. But notably they still have green pixels.

still have green pixels? RYYB - no green. Maybe you meant something else?

As far as I can tell, the spectral responses in Jack's post are for a Sharp Quatron LCD display, rather than for a CMOS or CCD sensor.
The Quatron display has R Y G B pixels. According to Thomas Bangert of QMUL (here) the Y response is simply a linear combination of the R+G outputs, and fails to offer the increase in gamut that a fourth yellow primary could provide.

 

[Eric Fossum]

Photo_Genius wrote: ... but with RYB, how do you mix the color green?

... So green = yellow - red. But notably they still have green pixels.

still have green pixels? RYYB - no green. Maybe you meant something else?

As far as I can tell, the spectral responses in Jack's post are for a Sharp Quatron LCD display, rather than for a CMOS or CCD sensor.

The Quatron display has R Y G B pixels. According to Thomas Bangert of QMUL (here) the Y response is simply a linear combination of the R+G outputs, and fails to offer the increase in gamut that a fourth yellow primary could provide.

Thanks Alan for the clarification and comment. Makes sense.

 

[chriswy]

Guys, DXO says P30 Pro uses a 1/1.7 inch sensor for its main camera, while others are using something close to 1/2.55 inch sensor... That itself counts for about 1x increase in sensor area... With larger sensor area, you can pack more powerful ADC into it. That means you get more than 100% performance increase out of 100% increase of area.

That 50% increase in light intake could be a numerical gimmick and not necessarily the main contributor to its boost in performance. But it definitely sounds cool compared with "Hey we have a BIGGER sensor".

However, color space conversion is a big headache if you moved away from RGB. Some early comparisons already shows P30 Pro's dailylight photos have less impressive color rendition compared with Samsung.

I am not a big fan in "Yellow = Green + Red" philosophy. It causes problems when ratio of Green over Red is close to extremes (large or small). Also, statistically this actually increases sampling error as you add another step in the calculation, and will cause issues when ratio of Green over red is close to extreme.

 

[Gareberry]

I mean, what use is this? Nothing here was taken in the dark, it's essentially as bright as day in all of these images and the writer admits they didn't take any dark shots. I saw a side by side of images taken in the dark, where human eyes can barely make out objects and phones can do diddly squat, the Huawei was extraordinary.

 

[Mikael Risedal]

I mean, what use is this? Nothing here was taken in the dark, it's essentially as bright as day in all of these images and the writer admits they didn't take any dark shots. I saw a side by side of images taken in the dark, where human eyes can barely make out objects and phones can do diddly squat, the Huawei was extraordinary.

well the Huawei in based on a sony sensor , so wha to expect

 

[preferredfault]

RYYB does and doesn't collect more light than other sensors. The aim is that with "yellow" filters, they collect a mesh of wavelengths of light. I.e. red and green, to get a better signal/brightness value for those pixels. But I would say that color accuracy would suffer a lot because you'd have no way of determining the ratio of red vs blue that the single pixel is picking up. Thus they wouldn't be able to tint the final output to account for both colors. So color accuracy would suffer a lot. It could however help with noise, but only in the sense of generating extra signal to fight against the noise generated by the DAC conversion of sensor signal to digital data. But the R and B pixels would be noisier, and that's also primarily where the color data is interpreted.

Demosiaicing works like this: If the average luminance value in an area for red pixels is highest, it flips all green and blue pixels to red, and the luminance values those pixels collects determines if it's brighter or darker red, i.e. the shade of red.

Does the same thing for blue and green pixels. And that's how you can take a picture of a red rose while only 25% of the pixels can see red light.

So you can imagine that if yellow pixels have the greatest luminance/signal strength because they collect more light, then demosaicing would lead to a lot of artifacts and color accuracy issues.